Skip to content
verify mcp Beta VerifyMCP is currently in beta. If you notice any issues, get in touch and we’ll put it right.

crit

PYPI · CRIT

Build, evaluate and serve a bespoke LLM inference engine for your own GPU with crit's commands.

Available components

−12 this week 63 Trust /100

Last scanned (13 hours ago)

Scores higher than 55% of PyPI servers

Category AI and Machine Learning

Trust breakdown (7 categories)

How this component scores in each security and reliability category. Every signal is checked automatically from public evidence about the published package, including repeated runs of it in an isolated sandbox, and we only credit what we can confirm. How we score → Why this is hard to score →

Supply Chain Security50
  • Malware scan not yet available for this package.Unverified
  • No known CVEs affecting this package version or its production dependencies.Pass
  • Runs setuptools.build_meta at install time, a recognised build step with no custom scripting around it. View diagnostics → Pass
  • 2 of 25 dependencies flagged as unhealthy. View diagnostics → Partial
Provenance & Transparency97
  • Source repository is publicly reachable at the declared URL. View diagnostics → Pass
  • Cryptographically verified build provenance (signed, bound to crit-labs/crit). View diagnostics → Pass
  • Clear OSI-approved license (Apache-2.0).Pass
  • Actively maintained (last published 2 days ago).Pass
  • Disclosure check failed: no security disclosure policy was found in the source repository. See how to fix → Fail
Schema Quality & AI Usability33
  • 0% of prompts and resources have a non-trivial description (not blank, and not just the item's name).Fail
  • AI-judged instruction clarity (good).Pass
  • Context-footprint check failed: tool/resource definitions use about 12725 tokens (~176/item across 72 items; 62 tools + 10 resources), over budget; trim descriptions and params. See how to fix → Fail
  • Usage-examples check failed: none of the tools include examples. See how to fix → Fail
Stability & Change Management40
  • Stability observed for 12 of 30 days with no destabilising changes; credit accrues until the full window elapses.Partial
Tool Coverage96
  • 100% of tools have a non-trivial description (not blank, and not just the tool's name).Pass
  • 88% of tool parameters carry a description.Partial
Tool Safety94
  • No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.Pass
  • 13 of 17 tool(s) whose name or description implies an irreversible operation declare an MCP destructiveHint annotation; "crit_eval_context" implies "eval" and declares readOnlyHint instead, contradicting what its own name says it does. See how to fix → Partial
  • An AI judge read all 64 captured unit(s) of tool text and found none that tries to manipulate the model reading it.Pass
Capabilities100
  • Implements a supported MCP spec version (2025-11-25); the latest is 2026-07-28.Pass
What we tested

Checks we ran (15)

  • Known CVEs
  • Install scripts
  • Dependency health
  • Source repository
  • Provenance
  • License
  • Maintenance
  • Security disclosure
  • Schema quality
  • Stability
  • Tool coverage
  • Injection markers
  • Destructive annotations
  • Judged manipulation
  • MCP protocol

Checks we could not complete (1)

These are gaps on our side, not findings about this package. We only credit what we can confirm, so a category made up entirely of them scores 0.

  • Malware scan in Supply Chain SecurityMalware scan not yet available for this package.

Does not apply to this package (2)

  • Endpoint SecurityScored on hosted endpoints only. A package has no live endpoint of its own for us to probe.
  • Transport & ReachabilityScored on hosted endpoints only. A package has no live endpoint of its own for us to probe.

The score does not measure everything. What the score does not measure →

Changelog

Every change we have recorded for this component, newest first. Security-relevant changes are always shown. ▲ marks a change for the better, ▼ a change for the worse; unmarked changes are neutral.

  • 20 Nov 26 +1

    No change was recorded against any check on this day. Stability & Change Management went from 33 to 37. That category is still filling its 30-day observation window: 10 days of observed history at the previous scan, 11 at this one. The score rises as the window fills, whether or not the server changes.

  • 18 Nov 26 +1

    No change was recorded against any check on this day. Stability & Change Management went from 27 to 30. That category is still filling its 30-day observation window: 8 days of observed history at the previous scan, 9 at this one. The score rises as the window fills, whether or not the server changes.

  • 17 Nov 26 −15
    • Malware scan: pass → unverified ▼ security
  • 16 Nov 26 +1

    No change was recorded against any check on this day. Stability & Change Management went from 20 to 23. That category is still filling its 30-day observation window: 6 days of observed history at the previous scan, 7 at this one. The score rises as the window fills, whether or not the server changes.

  • 14 Nov 26 +1

    No change was recorded against any check on this day. Stability & Change Management went from 13 to 17. That category is still filling its 30-day observation window: 4 days of observed history at the previous scan, 5 at this one. The score rises as the window fills, whether or not the server changes.

  • 12 Nov 26 +15
    • Malware scan: unverified → pass ▲ security
  • 11 Nov 26 −14
    • Malware scan: pass → unverified ▼ security
  • 9 Nov 26 73

    First indexed and scored.

Diagnostics

Diagnostic detail from the automated scan of this channel: what the scanner observed at each step, so you can see exactly where a check passed or failed. It is informational only and never changes the trust score.

Captured 21 Nov 2026 · Analysed pypi/crit@0.17.2

Provenance Verified

A signed build attestation was found and verified, binding this exact artifact to the source repository it claims to come from.

Result Verified
Ecosystem pypi
Reason Verified
Discovered via Registry attestation endpoint
Source repo crit-labs/crit
Certificate issuer https://token.actions.githubusercontent.com
Certificate SAN https://github.com/crit-labs/crit/.github/workflows/release.yml@refs/tags/v0.17.2
Rekor log index 3763503317
Predicate type PyPI publish attestation https://docs.pypi.org/attestations/publish/v1
Subject digest sha256:33e3a4022c60291a4978b5b22949695a6b395a456e03839987c3b1d343de1526

Background: How many MCP packages publish verified provenance →

Install scripts 1 script
Hook Tier Command
build_backend Known build tool setuptools.build_meta

Background: Why install scripts are a supply-chain risk →

Dependencies 25 packages
Packages resolved 25
Stale 1
No linked repository 1
Tree resolution Complete

Background: SBOMs and build attestations, explained →

MCP tools · 62 exposed · ~12,417 tokens

The tools this component advertises to a client, with an estimated token cost for each. Expand a tool to see its parameters and schema. The per-tool counts are indicative and are not scored directly; the schema's total context footprint is one signal in Schema Quality & AI Usability. A tool's description is untrusted text the model reads on every call, which is what makes this list a security surface and not just an inventory: how tool poisoning works →

Tool Tokens
crit_eval_ab ~262

Alternate two engines' speed runs in one session and report paired ratios

NameTypeReqDescription
adapterstring–Adapter of engine A
adapter_bstring–Adapter of engine B (default: engine A's adapter)
configstring–JSON configuration overrides of engine A (text or a file), over its packaged configuration.json
config_bstring–JSON configuration overrides of engine B (text or a file), over its packaged configuration.json
contextinteger–Context both engines are built for (the run's kv_tokens)
datastringyes–
modelstringyes–
outstringyesResult JSON path
packagestring–Engine A's crit package, opened through Core's loader
package_bstring–Engine B's crit package (default: engine A's engine)
repetitionsinteger–Measured repetitions per engine and round
roundsinteger–Rounds; each round runs both engines, order alternating
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)
warmupsinteger–Warmup runs per engine and round

No output schema declared.

No examples provided.

crit_eval_compare ~58

Print a side-by-side table of result files

NameTypeReqDescription
resultsarrayyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_context ~271

The max context for a model on a device: the model's positions and the KV cache that fits next to its weights

NameTypeReqDescription
context_reserve_gibnumber–Device memory kept out of the KV cache for workspaces and speed techniques (default: settings.toml context_reserve_gib, else the protocol's reserve_gib)
devicestringyesDevice file from `crit eval probe`
drafterarray–A companion drafter the engine may load beside the model; its stored weights are kept out of the KV cache (repeatable)
jsonboolean–Print the computation as JSON
modelstringyesModel directory
ram_headroom_gibnumber–Host memory left free, for the placement scenario (default: settings.toml ram_headroom_gib, else 2)
requested_contextinteger–The longest request wanted (default: settings.toml context_tokens, else 200,000)
vram_headroom_gibnumber–Device memory left free (default: settings.toml vram_headroom_gib, else 1)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_fetch ~169

Fetch the checkpoint's reference package (prepared data and reference scores) into the data, so a run needs only the engine

NameTypeReqDescription
chat_templatestring–Chat template (jinja) to render the prompts with, for a model that carries none (default: the one crit ships for the model's family); it becomes part of the model's identity; the package was prepared…
datastringyesData directory to write
modelstringyesThe checkpoint the engine serves
packagesstring–Package source: a folder or an https base URL (default: settings.toml's eval_packages)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_journal ~89

How far a round got: its journal's engines, parts and requests, replayed and live, and how it ended

NameTypeReqDescription
journalstringyesA round's journal (R.journal.jsonl beside its result R.json)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_kernels ~360

Trace every kernel of the speed cases (CUPTI, Linux) and rank them per phase by time, launch width and share of peak

NameTypeReqDescription
adapterstring–Python file exporting create_eval_engine(context)
backendstring–Device API the engine runs on: cuda (CUPTI) or hip (rocprofiler-sdk)
casearray–Trace only these speed cases (repeatable); default all
configstring–JSON configuration overrides for the adapter, text or a file, over the configuration.json packaged beside it
contextinteger–Context the engine is built for (the run's kv_tokens)
datastringyesPrepared suite directory for this model
devicestring–Device ceilings from `crit eval probe` for share of peak
hipccstring–hipcc of a ROCm that ships rocprofiler-sdk (--backend hip)
modelstringyesModel directory passed to the adapter
nvccstring–nvcc of a CUDA toolkit that ships CUPTI
outstringyesKernel report JSON path
packagestring–A crit engine package, opened through Core's loader
topinteger–Kernels listed per phase
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)
workstring–kernel_work rules ahead of the adapter's own, as JSON text or a file: [{"match": NAME_PART, "term": matmul|attention|recurrent|other|draft, "format": nvfp4|fp8|bf16, ...}]

No output schema declared.

No examples provided.

crit_eval_pack ~79

Add prepared data scored by the reference to a package folder

NameTypeReqDescription
datastringyesPrepared suite directory holding reference scores
packagesstringyesPackage folder to add it to
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_pin ~172

Record one engine's plain greedy output of every speed case as the data directory's pinned continuation (decode is then measured on that text)

NameTypeReqDescription
adapterstring–Python file exporting create_eval_engine(context)
configstring–JSON configuration overrides for the adapter, text or a file, over the configuration.json packaged beside it
contextinteger–Context the engine is built for (the run's kv_tokens)
datastringyesPrepared data directory to pin the continuation in
modelstringyesModel directory passed to the adapter
packagestring–A crit engine package, opened through Core's loader
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_precheck ~365

In minutes, the gates a candidate most often fails a whole round on: honesty of short replies and of one long reply, the request history around them, device memory and (with an adapter) the lower context

NameTypeReqDescription
adapterstring–Python file exporting create_eval_engine(context)
chat_templatestring–Chat template (jinja) to render the prompts with, for a model that carries none (default: the one crit ships for the model's family); it becomes part of the model's identity
configstring–JSON configuration overrides for the adapter
contextintegeryesContext the engine is built for (the run's kv_tokens)
datastringyesPrepared suite directory for this model
modelstringyesModel directory passed to the adapter
no_lowerboolean–Skip the lower-context reopen
outstring–Write the items as JSON here
packagestring–A crit engine package, opened through Core's loader
quality_modestring–The run's quality mode (default speed): speed bounds honesty as Eval does and lets history part at near-ties; preserve does neither
served_modelstring–With --url: the model to request
urlstring–Base URL of an OpenAI-compatible server the engine runs behind (no lower-context check: a server cannot be opened again)
vram_headroom_gibnumber–Device memory to find free after the load (default: settings.toml vram_headroom_gib, else 1)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_prepare ~179

Freeze the suite as token ids for one model

NameTypeReqDescription
chat_templatestring–Chat template (jinja) to render the prompts with, for a model that carries none (default: the one crit ships for the model's family); it becomes part of the model's identity
datastringyes–
modelstringyes–
samplingstring–The user's sampling for the sampled speed suite in place of the model's generation_config.json, e.g. temperature=0.6,top_p=0.95,top_k=20 (temperature required; top_p, top_k, min_p optional). Results…
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_probe ~214

Measure device bandwidth and GEMM rates per weight format (needs nvcc, or hipcc on ROCm)

NameTypeReqDescription
backendstring–Device API to probe (default: cuda when nvcc is found, else hip when hipcc is)
devicesstring–Probe several GPUs (CUDA): each one's figures and the copy rate between each pair, for multi-GPU limits (default: device 0 alone)
hipccstring–hipcc to build the HIP probe with (default: PATH, ROCM_PATH, /opt/rocm*)
nvccstring–nvcc to build the probe with (default: PATH, CUDA_HOME, /usr/local/cuda*)
outstringyesDevice JSON path for `run --device`
repeatsinteger–Probe runs; each value is the median
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_reference ~147

Score the prepared quality texts once on the checkpoint they were prepared from and store the scores under the data (torch + transformers, GPU; stop any engine first)

NameTypeReqDescription
allowanceboolean–Also score the texts with the activations quantized as the checkpoint declares and store how far that lands from the scores (allowance.json): what the format itself costs, reported, not judged; a sec…
datastringyesPrepared suite directory
modelstringyesThe checkpoint the data was prepared from
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_run ~1,266

Evaluate one engine (prepares the data first when missing)

NameTypeReqDescription
adapterstring–Python file exporting create_eval_engine(context)
adaptive_sampledstring–Repeat each sampled speed case at further seeds until its relative standard error reaches the protocol's target or its repetition cap (protocol sampled_speed.adaptive), recorded in the result; off by…
attribute_driftboolean–After quality, score it again on an engine opened with each of the adapter's exact_variants (one matrix group or arithmetic choice exact) and report each one's share of the drift budget
bandwidthnumber–Achievable device memory bandwidth in GB/s; overrides the --device value
baselinestring–Earlier result of the same data to show KPI changes against
budget_minutesnumber–The staged run's wall-clock budget, loads included (default: the protocol's 45)
certifyboolean–The candidate that will ship: stage 3 whatever the budget, then one request at the required context itself, which the capacity gate needs
chat_templatestring–Chat template (jinja) to render the prompts with, for a model that carries none (default: the one crit ships for the model's family); it becomes part of the model's identity; data prepared already mu…
configstring–JSON configuration overrides for the adapter, text or a file, over the configuration.json packaged beside it
contextinteger–Context (prompt plus output) every engine is built and measured for: the run's kv_tokens. Default: crit-eval's max context for the model on the --device
context_reserve_gibnumber–Device memory kept out of the KV cache when crit-eval computes the max context: activations, workspaces, graphs and speed techniques (default: settings.toml context_reserve_gib, else the protocol's r…
datastringyesPrepared suite directory for this model
devicestring–Device ceilings from `crit eval probe` (bandwidth and GEMM rates)
exactnessstring–reported (default): agreement with plain greedy is shown and the quality budget judges the output; required: any difference fails the gate
jitter_seedinteger–Draw the held-out prompts' and the long-context curve's lengths from this seed, a recorded result's jitter.seed, to measure at its lengths again (default: a new seed per run; a resumed round's from i…
keep_disturbedboolean–Finish the round even when a speed case's repetitions show a disturbed device (by default the round stops there; protocol speed.disturbed_spread)
labelstring–Name shown in comparisons
modelstringyesModel directory passed to the adapter
no_journalboolean–Write no journal beside the result (by default every engine opened and request served is one line of OUT's .journal.jsonl, synced as it completes)
onlyarray–Run only these parts (repeatable); default is all. context: the long-context curve and retrieval up to the required context and the speed cases again at a lower context; history: the same requests on…
outstringyesResult JSON path
packagestring–A crit engine package (plugins/ and configuration.json), which Core's eval-v1 adapter opens through Core's loader
profilestring–A protocol profile fixing what the run measures (protocol profiles). hub: crit-hub submissions, 200,000 tokens of context, the curve at its two ends, one engine load
quality_budgetstring–Drift limits over the protocol defaults, e.g. '{"ppl_ratio_max": 1.02}'; keys ppl_ratio_max, kld_mean_max, kld_p99_max
quality_modestring–The quality mode: how much output drift from the checkpoint the engine may trade for speed, whatever format the checkpoint is stored in; both modes judge against the same reference (its stored weight…
repetitionsinteger–Override the protocol's measured repetitions
requested_contextinteger–The longest request wanted, which caps the max context crit-eval computes without --context (default: settings.toml context_tokens, else 200,000)
resumestring–Go on from a round that stopped (interrupted, crashed, a disturbed device, a hung engine): its journal (R.journal.jsonl beside its result R.json) answers the requests it served, the engine loads only…
served_modelstring–With --url: the model to request (default: the one the server lists)
stagestring–How far a run without a profile measures (protocol stages): by default as far as the budget allows at the engine's measured rates; 1, 2 or 3 at most that far; full every part at the protocol's repeti…
urlstring–Base URL of an OpenAI-compatible server the engine runs behind, for example http://127.0.0.1:8000/v1 (needs --context: the context the server was started with)
vram_headroom_gibnumber–Device memory left free when crit-eval computes the max context (default 1)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)
warmupsinteger–Override the protocol's warmup runs

No output schema declared.

No examples provided.

crit_eval_status ~188

Gates, goals, KPIs against their physical limits, gaps and missing inputs

NameTypeReqDescription
baselinestring–Earlier result to show KPI changes against
datastring–The data directory the result was measured on: its model arithmetic prices prefill for a result that does not record it
devicestring–Price prefill and time to first token with this `crit eval probe` result's rates instead of the result's own device
jsonboolean–Print the status as JSON
kernelsstring–A kernel trace of the engine (crit eval kernels): its prefill GPU time per floor term against the term's floor, with a gap each
resultstringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_eval_workbench ~205

Check, time or profile a kernel on its workbench folder (crit kernels new): reference, floor, yardstick

NameTypeReqDescription
actionstringyes–
devicestring–bench: a `crit eval probe` result, for the floor
folderstringyesThe workbench folder
itersinteger–bench: timed calls per shape and column count
ncustring–profile: Nsight Compute's ncu, to add its sections
outstring–check and bench: the result JSON (default FOLDER/check.json, bench.json)
pythonstring–The engine's Python (torch and the GPU); default this one
shapeinteger–One shape of kernel.json by index (profile: default the first)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_inference_check ~51

Check the techniques' code against their metadata

NameTypeReqDescription
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_inference_list ~55

List the techniques by category

NameTypeReqDescription
jsonboolean––
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_inference_search ~67

The techniques whose metadata holds every word of the query

NameTypeReqDescription
jsonboolean––
querystringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_inference_show ~65

One technique's guidance, symbols and files

NameTypeReqDescription
jsonboolean––
querystringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_job_output ~79

A job's command output (its JSON result) or errors (progress and reports) from a byte offset; offset -1 reads the end

NameTypeReqDescription
jobstringyesThe job a command's call returned, such as job-3
limitinteger––
offsetinteger––
streamstring––

No output schema declared.

No examples provided.

crit_job_stop ~76

Interrupt a running job as Ctrl+C would (crit releases its leases and settles what it can), ending its process tree after grace_seconds; an Eval stopped so goes on with crit_run_eval resume

NameTypeReqDescription
grace_secondsnumber––
jobstringyesThe job a command's call returned, such as job-3

No output schema declared.

No examples provided.

crit_job_wait ~77

Wait for a job to end, at most SECONDS; returns its result once it ended, else its state and newest errors

NameTypeReqDescription
jobstringyesThe job a command's call returned, such as job-3
secondsnumber–Longest wait (at most 3600); call again to go on

No output schema declared.

No examples provided.

crit_jobs ~20

This server's jobs: state, exit code, logs

Input schema present but exposes no named parameters.

No output schema declared.

No examples provided.

crit_kernels_inspect ~67

Explain every declared applicability match, rejection or unknown

NameTypeReqDescription
catalogstringyes–
requeststringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_kernels_inventory ~57

Hash local artifact files without importing them

NameTypeReqDescription
rootstringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_kernels_new ~377

Scaffold a workbench folder for a kernel: reference, shapes, yardstick, harness (measure it with crit eval workbench)

NameTypeReqDescription
column_invariantstring––
columnsstring–column (token) counts, comma-separated
formatstring–bf16, fp8 or nvfp4 (NVFP4 weights against bf16 activations) for gemv and gemm
fromstring–a kernel-builder project to start src/ from (a crit-kernels project, a Hub kernel's source); without it, kernel-builder init starts one
modelstring–take the shapes from this checkpoint's matrices in the format
namestringyes–
opstringyesgemv or gemm (a working start in bf16, fp8 or nvfp4), or any other operation name for a harness to fill in
repeatablestring–no: the kernel may give other bits on a repeat (atomics, a split the workload chooses); the check judges repeats within the tolerance. For quality mode speed engines
rootstring––
runstring–the run folder, for --starter
shapearray–N,K (repeatable)
starterboolean–the kernel --from copies is the starter (crit-kernels' workbench-starter: a simple, correct gemv/gemm in bf16 and fp8); allowed from a run's fourth candidate on
tolerancenumber––
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)
yardstickstring–the provider to time beside the kernel (provider syntax of plan levers)

No output schema declared.

No examples provided.

crit_kernels_retain ~104

Copy and verify a selected dependency into a new directory

NameTypeReqDescription
catalogstringyes–
destinationstringyes–
max_bytesinteger––
reasonstringyes–
requeststringyes–
rootstringyes–
selectstringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_kernels_search ~61

Search local implementation declarations

NameTypeReqDescription
catalogstringyes–
querystringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_kernels_target ~52

Observe host OS/architecture; ABI remains explicit

NameTypeReqDescription
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_kernels_verify ~59

Recheck a retained dependency's bytes and applicability

NameTypeReqDescription
directorystringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_models_drafter ~161

Distill a draft head for speculative decoding from the target itself and write its models_transform record

NameTypeReqDescription
calibrationstringyesJSONL of {"text": ...} documents unrelated to Eval
devicestring–torch device (default: cuda when available, else cpu)
outstringyesNew or empty directory for the drafter
recipestringyesmodels_recipe JSON with method drafter
recordstringyesNew models_transform JSON outside the source and output
sourcestringyesThe target model directory (the checkpoint the engine serves)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_models_eval ~126

Bind the Eval result of an engine serving the output to its record, with its quality replies scored on the source checkpoint

NameTypeReqDescription
datastringyesEval data directory prepared from the source checkpoint
evaluationstringyesNew models_evaluation JSON
recordstringyesmodels_transform JSON
resultstringyescrit eval run --url result of the engine serving the output; stop the server first
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_models_plan ~92

Resolve a recipe against a checkpoint's headers: targets and bytes

NameTypeReqDescription
jsonboolean–Print the resolved recipe and plan as JSON
recipestringyesmodels_recipe JSON
sourcestringyesUnquantized source model directory
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_models_ptq ~149

Quantize a checkpoint and write its models_transform record

NameTypeReqDescription
calibrationstring–JSONL of {"text": ...} documents unrelated to Eval (gptq only)
devicestring–torch device (default: cuda when available, else cpu)
outstringyesNew or empty directory for the transformed checkpoint
recipestringyesmodels_recipe JSON
recordstringyesNew models_transform JSON outside the source and output
sourcestringyesUnquantized source model directory
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_models_verify ~61

Re-hash a transform's output and source against its record

NameTypeReqDescription
recordstringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_pull ~645

crit pull OWNER/NAME: install a crit-hub engine, so that crit serve --engine OWNER/NAME serves it. It reads the engine's card (from a crit-hub checkout with --hub, else crit-hub's index.json), checks the card's requirements against this machine (a crit eval probe result, and nvcc), fetches the engine's source at the commit the card pins (from a crit-engines checkout with --engines, else its repository), checks its configuration against the sealed run's in card.lock.json, fetches the checkpoint at its pinned Hugging Face revision (or takes a local folder with --model) and checks every file against card.lock.json, fetches and checks the card's drafter downloads the same way, runs the card's build command (or unpacks a prebuilt artifact whose toolchain matches this machine instead), and installs the built package in the engine store (crit.pull.store). Nothing is installed until every step has passed.

NameTypeReqDescription
build_from_sourceboolean–Build even when one of the card's prebuilt artifacts matches this machine
build_timeoutnumber–Seconds the build may take (default: 1800)
checkpointstring–The checkpoint, when the engine was measured on several
contextinteger–Context the engine serves at (default: the context its card measured)
devicestring–A crit eval probe result to check the requirements against (default: settings.toml's device, else a probe run now)
downloadarray–A local copy of one of the card's downloads (a drafter's folder, an artifact's file), checked like a download (repeatable; default: fetch it from Hugging Face)
enginestring–The crit-hub engine, as OWNER/NAME
enginesstring–Take the source from this crit-engines checkout, which holds the pinned commit (default: fetch that commit from the card's source repository)
forceboolean–Install again over an installed engine
hubstring–Read the card from this crit-hub checkout (default: crit-hub's index.json)
ignore_requirementsboolean–Install even when this machine does not meet the card's requirements
indexstring–crit-hub's index.json, a URL or file (default: https://crit-hub.dev/index.json)
listboolean–List the installed engines instead
modelstring–A local copy of the checkpoint, checked file by file and hard-linked into the store (default: download it at the pinned Hugging Face revision)
projectstring–crit checkout whose settings.toml gives the device and headroom and whose engines/ is the default store (default: this checkout); a crit installed without a checkout takes any folder (default: the cu…
storestring–Engine store (default: PROJECT/engines)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_ablate ~277

Serve a candidate built from plugins with each variant of its plugins (as built, at its base, without each and with each alone): a short bench of the run's cases and crit smoke per variant, and what each plugin adds, costs or breaks (docs/PLUGINS.md)

NameTypeReqDescription
candidatestringyes–
idstringyesExisting run identity
no_smokeboolean–Leave out crit smoke on each variant
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
repetitionsinteger–Repetitions per case and variant (default 3)
rootstring–Run root; defaults to PROJECT/runs
secondsnumber–Reservation length (default: the check process cap)
smoke_testarray–Run only this crit smoke test on each variant (repeatable)
variantarray–Serve only this variant (repeatable), e.g. full, base, without:ID, only:ID; crit plugins variants lists them
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_benchmark ~152

Append a case's benchmark evidence to a candidate

NameTypeReqDescription
candidatestringyes–
idstringyesExisting run identity
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
recordobject|array|stringyesA JSON object (written to a file for the command) or a JSON file's path
rootstring–Run root; defaults to PROJECT/runs
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_build ~252

Build a candidate's retained source with the recipe's build.command under a build attempt (log kept, attempt settled), then package the result

NameTypeReqDescription
candidatestringyes–
droparray–Remove these build-folder paths (intermediates) before packaging; repeatable
idstringyesExisting run identity
no_packageboolean–Build and settle only
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
recordobject|array|stringyesRecipe: build {toolchain, command[, dependencies]}; the build's outputs are bound as dependencies, and the package is the plugin package Core's loader runs. A JSON object (written to a file for the c…
rootstring–Run root; defaults to PROJECT/runs
secondsnumber–Reservation length (default: the build process cap)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_candidate ~169

Allocate a candidate ID and retain its source and proposal

NameTypeReqDescription
idstringyesExisting run identity
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
recordobject|array|stringyeslabel, rationale, strategy, configuration, store_decisions, parent/origin IDs. A JSON object (written to a file for the command) or a JSON file's path
rootstring–Run root; defaults to PROJECT/runs
sourcestringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_capture ~255

Write a case's chat request (captured/CASE/request.json) that renders to the workload's prompt tokens with the template variables candidates serve

NameTypeReqDescription
casestring––
idstringyesExisting run identity
inputsstring–Set the case's manifest row in this resolve payload (created when missing)
instructionstring–Line after the text (default: a summary request; '' for none)
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
rootstring–Run root; defaults to PROJECT/runs
systemstring–A leading system message
template_kwargsstring–chat_template_kwargs the request renders with, over recon's frozen template_variables (text or a file)
textstring–File of the user message's text (default: a neutral passage); it is cut to whole words and padded to the exact count
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_check ~195

Append a case's check evidence to a candidate

NameTypeReqDescription
candidatestringyes–
from_evalboolean|string–Instead of --record: record the arithmetic, functional and resources check of every case from the candidate's retained Eval result (default its newest). true gives --from-eval alone
idstringyesExisting run identity
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
recordobject|array|string–A JSON object (written to a file for the command) or a JSON file's path
rootstring–Run root; defaults to PROJECT/runs
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_cost ~246

Account the run's build cost from the agent's transcripts: wall time, tokens and API price per candidate, written to cost.json

NameTypeReqDescription
agentstring–The agent and model that built the run, as the record names it
idstringyesExisting run identity
pricesstring–JSON object or file of {model: {input, cache_read, cache_write_5m, cache_write_1h, output}} in USD per million tokens, over the list prices
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
rootstring–Run root; defaults to PROJECT/runs
transcriptarrayyesA Claude Code session, Codex rollout or plain usage JSONL (repeatable)
untilstring–Count requests up to this UTC timestamp (default: the run's stop, else now)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_create ~187

Create a run and its first segment from a request

NameTypeReqDescription
from_enginestring–Start from a crit-hub engine: its engines/<owner>/<name>/ folder (card.yaml, card.lock.json, evidence/) is recorded as the run's origin_engine
idstring–Existing run identity
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
recordobject|array|stringyesA JSON object (written to a file for the command) or a JSON file's path
rootstring–Run root; defaults to PROJECT/runs
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_eval ~469

Run Eval on a packaged candidate through crit serve, quality against the run's model's scores of the fixed texts (scored once into the data)

NameTypeReqDescription
baselinestring–Earlier Eval result to show KPI changes against
budget_minutesnumber–Eval's wall-clock budget, loads included (default: crit-eval's 45)
candidatestringyes–
certifyboolean–The candidate that will ship: stage 3 whatever the budget, then one request at the run's kv_tokens, which the resources check needs. On the candidate ladder the full Eval of candidate 5 on and of the…
datastringyesEval data directory prepared from the run's model
idstringyesExisting run identity
labelstring––
onlyarray–Run only these Eval parts (repeatable); default all
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
resumeboolean|string–Go on from an Eval attempt that stopped (interrupted, failed, timed out, a disturbed or hung engine): its journal's requests are answered again, the engine loads for the requests left; default the ca…
rootstring–Run root; defaults to PROJECT/runs
secondsnumber–Reservation length (default: to the segment's deadline; each request is bounded by the rates the engine measured)
stagestring–How far Eval measures (crit-eval's protocol stages): by default as far as its budget allows at the candidate's measured rates; 1, 2 or 3 at most that far; full every part as before stages. On the can…
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_measure ~203

Benchmark the run's cases on a packaged candidate through crit serve, once its checks are recorded (check --from-eval)

NameTypeReqDescription
candidatestringyes–
exploratoryboolean–Measure even when the candidate's checks would leave its benchmarks exploratory (not recorded, or a required one did not pass), for telemetry or memory: a benchmark that does not qualify keeps the ca…
idstringyesExisting run identity
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
rootstring–Run root; defaults to PROJECT/runs
secondsnumber–Reservation length (default: the benchmark process cap)
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_package ~169

Package a candidate's runnable files and manifest

NameTypeReqDescription
candidatestringyes–
idstringyesExisting run identity
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
recordobject|array|stringyesRecipe: build {toolchain, command[, dependencies]}. A JSON object (written to a file for the command) or a JSON file's path
rootstring–Run root; defaults to PROJECT/runs
sourcestringyes–
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

crit_run_plan ~170

Record the next candidate's plan

NameTypeReqDescription
checkboolean–Run the plan's checks and print the record crit would store, recording nothing (the plan command of a replay)
idstringyesExisting run identity
projectstring–crit checkout the run follows, whose settings.toml it reads (default: this checkout); a crit installed without a checkout takes any folder (default: the current one)
recordobject|array|stringyesA JSON object (written to a file for the command) or a JSON file's path
rootstring–Run root; defaults to PROJECT/runs
wait_secondsnumber–Seconds to wait for the command to end before returning its job to wait on with crit_job_wait (the command keeps running)

No output schema declared.

No examples provided.

Install

How do I install the crit MCP server?

crit runs locally as a PyPI package, launched with uvx crit mcp. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation.

pypi · crit

# add to Claude Code
claude mcp add crit-labs-crit -- uvx crit mcp
Common questions

What is the crit MCP server?

crit is an MCP server listed in the public MCP registry as io.github.crit-labs/crit. Build, evaluate and serve a bespoke LLM inference engine for your own GPU with crit's commands. This page covers its PyPI package (crit).

Is the crit MCP server safe to use?

crit scores 63 out of 100 on VerifyMCP. We found no known CVEs affecting it as of 21 November 2026. Its build provenance is signed and verified. That is a record of what we were able to check automatically, not an endorsement. The category breakdown on this page shows every signal behind the number, including the ones we could not confirm.

What tools does the crit MCP server expose?

crit exposes 62 tools: crit_run_create, crit_run_resolve, crit_run_plan, crit_run_check, crit_run_benchmark, and 57 more. Their descriptions and schemas cost roughly 12,417 tokens of context every time the server is loaded.

Is the crit MCP server still maintained?

crit is still listed as active in the MCP registry. We last reached this channel on 21 November 2026. Those dates come from our own scans of the registry and the channel itself, not from anything the publisher announced.

What licence is the crit MCP server under?

crit declares the Apache-2.0 licence, which is OSI-approved. That covers the source only, and says nothing about the cost of any service it calls.

What did VerifyMCP test on the crit MCP server?

VerifyMCP ran 15 checks on crit, covering Known CVEs, Install scripts, Dependency health, Source repository and 11 more. One item could not be completed, a gap on our side rather than a finding about crit: Malware scan. Endpoint Security and Transport & Reachability do not apply to a PyPI package, so they were not tested. The score does not measure everything, and the scoring documentation lists what it leaves out.