NoData
NPM · @NODATACHAT/MCP · 2 COMPONENTS
Governed access to sensitive data, secrets and APIs for AI agents. A signed proof for each decision.
Available components
Last scanned (20 hours ago)
Scores higher than 32% of npm servers
Category Security
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 Security98
- No malware found by supply-chain analysis.Pass
- No known CVEs affecting this package version or its production dependencies.Pass
- No install/post-install scripts declared.Pass
- 31 of 102 dependencies flagged as unhealthy. View diagnostics → Partial
Provenance & Transparency6
- Repository check failed: no source repository is declared. See how to fix → View diagnostics → Fail
- Provenance check failed: no build-provenance attestation is published. See how to fix → View diagnostics → Fail
- License check failed: the license (FSL-1.1-ALv2) isn't a recognized OSI-approved license. See how to fix → Fail
- Actively maintained (last published 1 days ago).Pass
- Security-disclosure policy not yet verified: we couldn't inspect the source repository.Unverified
Schema Quality & AI Usability58
- AI-judged instruction clarity (excellent).Pass
- Context-footprint check failed: tool/resource definitions use about 8432 tokens (~562/item across 15 items; 15 tools + 0 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 Management27
- Stability observed for 8 of 30 days with no destabilising changes; credit accrues until the full window elapses.Partial
Tool Coverage99
- 100% of tools have a non-trivial description (not blank, and not just the tool's name).Pass
- 97% of tool parameters carry a description.Partial
Tool Safety100
- No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.Pass
- We read all 15 captured tool definition(s), and no name or description among them implies an irreversible operation.Pass
- An AI judge read all 16 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
Checks we ran (15)
- Malware scan
- Known CVEs
- Install scripts
- Dependency health
- Source repository
- Provenance
- License
- Maintenance
- 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.
- Security disclosure in Provenance & TransparencySecurity-disclosure policy not yet verified: we couldn't inspect the source repository.
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 →
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.
- 21 Nov 26 +1
No change was recorded against any check on this day. Stability & Change Management went from 23 to 27. That category is still filling its 30-day observation window: 7 days of observed history at the previous scan, 8 at this one. The score rises as the window fills, whether or not the server changes.
- 20 Nov 26 0
- Package version: 1.6.65 → 1.6.66 functional
- 19 Nov 26 0
- Tool safety: pass → unverified ▼ security
- Stability: 0.17 → unverified ▼ security
- Capabilities: pass → unverified ▼ functional
- Tool coverage: 100 → unverified ▼ functional
- First check of Schema quality: unverified functional
- Package version: 1.6.64 → 1.6.65 functional
- 18 Nov 26 +16
- Malware scan: unverified → pass ▲ security
- Package version: 1.6.63 → 1.6.64 functional
- 17 Nov 26 −4
- Malware scan: pass → unverified ▼ security
- Known CVEs: unverified → pass ▲ security
- Dependency health: unverified → 0.88 ▲ functional
- Package version: 1.6.60 → 1.6.63 functional
- Package version: 1.6.60 → 1.6.62 functional
- Package version: 1.6.60 → 1.6.61 functional
- 16 Nov 26 +5
- Known CVEs: pass → unverified ▼ security
- Stability: 0.07 → unverified ▼ security
- Tool safety: pass → unverified ▼ security
- Malware scan: unverified → pass ▲ security
- Dependency health: 0.88 → unverified ▼ functional
- Capabilities: pass → unverified ▼ functional
- Tool coverage: 100 → unverified ▼ functional
- First check of Schema quality: unverified functional
- Package version: 1.6.50 → 1.6.60 functional
- Package version: 1.6.50 → 1.6.58 functional
- Package version: 1.6.50 → 1.6.57 functional
- 15 Nov 26 +11
- Known CVEs: unverified → pass ▲ security
- Malware scan: unverified → pass ▲ security
- Dependency health: unverified → 0.88 ▲ functional
- Package version: 1.6.45 → 1.6.50 functional
- Package version: 1.6.45 → 1.6.47 functional
- Package version: 1.6.45 → 1.6.46 functional
- 14 Nov 26 −25
- Known CVEs: pass → unverified ▼ security
- Malware scan: pass → unverified ▼ security
- Dependency health: 0.88 → unverified ▼ functional
- Package version: 1.6.43 → 1.6.45 functional
- Package version: 1.6.43 → 1.6.44 functional
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 npm/@nodatachat/mcp@1.6.66
Provenance No attestation
The registry publishes no build provenance for this version, so there is nothing to verify.
| Result | No attestation |
|---|---|
| Ecosystem | npm |
Background: How many MCP packages publish verified provenance →
Dependencies 102 packages
| Packages resolved | 102 |
|---|---|
| Stale | 31 |
| Tree resolution | Complete |
Background: SBOMs and build attestations, explained →
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 →
nodata_choose Choose what crosses from a folder ~518
CHOOSE WHAT CROSSES from a folder, per kind of detail the last nodata_scan_folder found (ID numbers, names, emails, bank details, secrets, the user's own terms, files not read): each kind crosses or is withheld. A part of the folder holding a withheld kind goes out only as a clean copy, or stays closed when very sensitive; the rest goes as the folder is. The same choice as the site's "Choose what crosses" panel. Without changes it shows the panel ("May cross · N of M"): show it to the user as it is and let THEM choose. cross / withhold change kinds (by name, as shown); all:'cross' selects all, all:'withhold' withholds all. approve:true makes it policy: this computer's owner approves it in the browser (opened for them), then it is recorded in a signed receipt and applied as rules on the rules board. You only propose: never approve for the user, and call approve only when the user asked for it. Counts and kinds only, never a value. An approved policy puts the folder on AUTOPILOT: the read gate answers by the policy instead of asking on each file (a kind it never saw still asks). autopilot:false takes the folder off it. coverage:true answers "is everything covered, and is there anything new to enforce?" for every project and scanned folder on this computer: new projects not scanned yet, folders without a policy, files changed since the scan, kinds the policy never saw, and the one next step for each. Use it when the user starts a new project, adds a folder or files, or asks whether they are protected.
| Name | Type | Req | Description |
|---|---|---|---|
| all | string | – | Select all, or withhold all. |
| approval_id | string | – | An approval already asked for (from a WAITING answer). |
| approve | boolean | – | true = ask this computer's owner to approve the choice as policy, in the browser. Only when the user asked. |
| autopilot | boolean | – | With approve: false approves without autopilot (default on). Alone: false takes the folder off autopilot. |
| coverage | boolean | – | true = the coverage report for every project and scanned folder on this computer. |
| cross | array | – | Kinds that may cross (names as the panel shows them). |
| path | string | – | The folder, as scanned with nodata_scan_folder. Not needed with coverage:true. |
| withhold | array | – | Kinds to withhold. |
No output schema declared.
No examples provided.
nodata_clean_folder Build a verified clean folder next to it ~331
DELIVER a clean folder: builds "<folder> (clean)" NEXT to the user's folder and changes nothing in theirs. A file with nothing sensitive is copied as it is; a file with findings goes in as its verified clean copy (values replaced with tokens, a workbook stays a workbook); a scanned PDF goes in as a PDF with its values and the user's terms painted black on the page images (checked by reading it again with OCR); a file NoData cannot vouch for (not read, PDF readers disagreed, a picture read by OCR, holds one of the user's terms) is held back and named with why. The clean folder is then scanned again by the same scan and anything still found is taken out. Call it right after nodata_scan_folder when the verdict is NOT READY or NOT PROVEN, without asking: it is the first thing the user gets. Lead your reply with the folder, then ask the one question its answer gives. Calling it again rebuilds the folder (after new terms or the amounts dial). lock_originals:true also builds "<folder> (locked)": every file the clean copy replaced or held back, sealed whole (the nodata_protect seal) so the full original opens only for its owner, with a receipt. Nothing in the user's folder changes and nothing is deleted. Use it when the user wants the clean copy AND an original they can still open.
| Name | Type | Req | Description |
|---|---|---|---|
| lock_originals | boolean | – | Also make a locked copy of the sensitive originals next to it. Default false. |
| path | string | yes | The folder to make a clean copy of. |
No output schema declared.
No examples provided.
nodata_connect Connect this computer to your organisation ~214
Connect this computer to the user's NoData organisation. Call with NO arguments: it opens a page in the browser where the user signs in (or creates a free account) and clicks Approve — nothing is copied; then /mcp. Other ways, only if that cannot work: step:"key" opens a private window for an org key (never typed into the chat); step:"signup" only opens the signup page. Never ask the user to paste a key into the chat.
| Name | Type | Req | Description |
|---|---|---|---|
| api_key | string | – | LEGACY — only if the user already pasted it; prefer step:"key", which keeps the key out of the chat. |
| step | string | – | Omit (default): approve this computer in the browser — a computer connected before reconnects under the same identity. new: connect as a NEW computer (e.g. to move it to another organisation; files i… |
No output schema declared.
No examples provided.
nodata_get_started Get started with NoData ~160
Call this FIRST (no arguments). NoData in one line: what you can achieve — an agent that cannot leak, a look at what is already exposed, a secret we cannot read — with the single next step for each, and how to recover if the tools or MCP seem stuck. If the user asks what NoData can do, how to start, what is missing, which version runs, or why the tools are missing, call this and relay it — it re-checks this machine each time. For ANY question about NoData pass it as `question`: the answer comes from NoData's answer bank and is recorded — answer only from it.
| Name | Type | Req | Description |
|---|---|---|---|
| question | string | – | The user's question about NoData, as they asked it. |
No output schema declared.
No examples provided.
nodata_node Register this computer as an organisation node ~172
Register THIS MACHINE as a Capsule node of the organisation — so "this machine is guarded" is something the organisation can verify, not only the user. One node per machine; every agent here (Claude Code, Cursor …) is a workload on it; billing stays per decision, a node is not a billing unit. Uses the Capsule agent's own enrolment and signed heartbeat. The heartbeat reports only: a machine id (a salted hash, never the host name), the agents' names, whether the read gate is on, and the accepted terms version — never content. ASKS FIRST: without confirm:true it shows the status, or what registering would send. Call with confirm:true only after the user approves.
| Name | Type | Req | Description |
|---|---|---|---|
| confirm | boolean | – | true = register this machine. Omit to see status / preview. |
No output schema declared.
No examples provided.
nodata_open Open a sealed file ~225
RELEASE a sealed file: open <file>.ndc back to its original next to it (never overwrites), and check its hash matches what was sealed. NoData hands the key half back only to the device the seal belongs to, or to another computer of the same person it was sealed for (each opens its own copy with its own key), in the same organisation, after a signed proof, and never after a revoke. A PERSON approves every opening: the call asks NoData, a page opens in the browser, and the file opens only after someone signed in to the organisation clicks Approve (the agent cannot approve; one approval opens once, within 5 minutes). If the call returns WAITING FOR A PERSON, tell the user to approve, then call again. Refuses when this computer is not connected. ASKS FIRST: without confirm:true it only says what would be opened. Writes a local receipt.
| Name | Type | Req | Description |
|---|---|---|---|
| confirm | boolean | – | true = open it. Omit to preview. |
| path | string | yes | The .ndc file to open. |
No output schema declared.
No examples provided.
nodata_plan Recommend and apply a file policy ~380
RECOMMEND a policy after a scan, per file, and apply it only once the user approves. For every file with findings it proposes a tier (Restricted / Confidential / Internal) and an action: "both" (clean copy to work on + lock the original), "protect" (lock), "redact" (clean copy) or "keep"; credentials also get a rotate-at-source step. ASKS FIRST: without confirm:true it only shows the plan. The user may change any action — pass them as `actions` {"relative/path": "protect"|"redact"|"both"|"keep"} — then call again with the same path, the same actions and confirm:true. Proposing is not accepting: never confirm on the user's behalf. Content-free: paths and counts only. SCOPE: this decides the files on this computer only; what may cross to a partner or an AI agent is nodata_choose, a separate decision this plan does not set. An approved plan runs in ROUNDS of up to 50 files: when the result says CONTINUE, call again with exactly the same path, actions and confirm:true (the user already approved; do not ask again) until it says ALL DONE. Finished files are never redone.
| Name | Type | Req | Description |
|---|---|---|---|
| actions | object | – | The user's changes: by relative path, or by folder / pattern for many files at once ("scripts/**": "keep", ".next/**": "keep", "*.env": "protect"). An exact path beats a pattern; a more specific patt… |
| confirm | boolean | – | true = the user approved this plan. Omit to show it. |
| path | string | yes | Folder or file to plan for. |
| remove_originals | boolean | – | When locking, delete the plaintext original. Default false (originals kept). |
No output schema declared.
No examples provided.
nodata_protect Seal files on this computer ~313
LOCK files: seal each one as <file>.ndc (AES-256-GCM, sealed to this device's key, which never leaves this machine). Every seal belongs to someone: it is registered at NoData under this computer's approved identity, NoData keeps only a key half it cannot open, and every opening is receipted and revocable. Refuses when this computer is not connected (nodata_connect): there is no anonymous local seal. Given a folder, seals only the files the scan finds exposure in; given a file, seals that file. ASKS FIRST: without confirm:true it only lists what would be sealed and changes nothing — show that list to the user and call again with confirm:true only after they approve. Writes a local receipt (hashes, seal ids) under <folder>/.nodata/receipts. remove:true also deletes the plaintext originals. For a file ONLY the person should hold (Holder Lock: their own code opens it; the organisation recovers it only through a request, an approver and their window to refuse), tell them to run `npx -y @nodatachat/mcp@latest hold <file>` in their own terminal: the code is typed hidden there. Never ask for that code in the chat.
| Name | Type | Req | Description |
|---|---|---|---|
| confirm | boolean | – | true = do it. Omit to preview. |
| path | string | yes | File or folder to protect. |
| remove | boolean | – | Delete the plaintext originals after sealing. Default false. |
No output schema declared.
No examples provided.
nodata_redact Make a clean copy without personal values ~504
WORK on the rest: make a clean copy of a file with every sensitive value replaced by a typed token ([CARD], [ID], [IBAN], [PHONE], [EMAIL]); in CSV/TSV and Excel a column whose header names personal data (name, address, account number…) is replaced whole with one stable token per value ([NAME-1], [NAME-2] … — the same person keeps the same token), so rows and columns still line up and still group by person. Columns of amounts stay whole. Same detectors as the scan. The copy keeps the file's type: in Excel (.xlsx) the same column rule applies per sheet and the copy stays a workbook with its formatting; a Word document (.docx) or a deck (.pptx) stays one, with its tables, headings and slides, and its pictures blanked. The original is untouched; the copy is written next to it as <name>.clean.<ext> (a PDF with a text layer, or a document with charts or embedded files → .clean.txt, and the answer says why). A SCANNED PDF (no text layer) stays a PDF: every page is read by OCR on this computer, each value the detectors find and each of the user's terms (nodata_scan_terms) is painted black on the page image, and the copy is a new PDF of page images only, checked by reading it again with OCR. Never black out a PDF with a script of your own: this is the copy that carries the proof. The answer holds counts, kinds, the path and the proof, never the document's text: work from the clean copy at that path. ON SCREEN: someone may be sitting next to the user. Answer with totals, counts and summaries by default; ask before printing rows or individual values, even from the clean copy (a value the detectors did not recognise is still in it).
| Name | Type | Req | Description |
|---|---|---|---|
| near_miss | boolean | – | Scanned PDF only: also paint a word one letter off one of the user's terms, or behind a Hebrew prefix. Off by default: it also paints ordinary words one letter away. |
| path | string | yes | The file to redact (csv, tsv, txt, md, json, log, docx, xlsx, pptx, pdf). |
| return_text | boolean | – | Ignored: the text is never returned. Read the clean copy at the path in the answer. |
No output schema declared.
No examples provided.
nodata_rules Set what agents may do with files ~1,809
The RULES BOARD for files, folders and web sites on this computer: what an AI agent may do with each one — allowed, ask the user, clean copy only, or not allowed — optionally only for certain agents (claude-code, cursor …) or sub-agents, at certain times, or for certain actions (read, search, run, write; on a site: browse, interact) or file types. A web site rule (site:'example.com', its subdomains too; site:'*' = every site) holds an agent that drives a browser over MCP or fetches a page: browse = open or read a page, interact = click, type, submit. e.g. {action:'add', site:'bank.example.com', effect:'ask', actions:'interact'} = the user approves every click there (the human checkpoint: a CAPTCHA, a login, a payment is the user's to pass, never the agent's). An allowed site is also a TRUSTED DESTINATION: after the session read a web page, an API answer or a partner's command, a network step waits for a person, except a curl / wget whose every address is a named allowed site (https, no proxy, no -k; site:'*' does not count). e.g. {action:'add', site:'api.elevenlabs.io', effect:'allow', actions:'run'} trusts only commands to it. A TRUSTED MCP SERVER is the same for another server's tools: after outside content, a tool that writes through another MCP server waits for a person, except the tools the owner named: {action:'add', server:'dreamz', tools:['render_take','take_status'], effect:'allow'} (server as in mcp__<server>__<tool>; each tool named exactly, no wildcard, never the whole server; a tool that sends a message to people, send, reply, forward, email, message, post, comment, invite, share, publish, is never trusted; agent, hours, days and origin may scope it). Several rules per file or folder; the most specific path decides, and among its rules the first whose conditions hold. Enforced by NoData's read gate on every read, search, command and write, and on every page an agent opens or acts on through a browser tool or WebFetch, whatever the agent. Plain fields,…
| Name | Type | Req | Description |
|---|---|---|---|
| action | string | yes | – |
| actions | – | – | add: which actions — on files: read, search, run, write; on a site: browse, interact. "write", ["read","search"], or "all but read". Omit = every action. |
| agent | – | – | add: which agent the rule is for: "cursor", ["cursor","windsurf"], or "all but claude-code". Omit = every agent. |
| approval | string | – | The code the user approved (loosening only). |
| checkpoint | string | – | approve: the id the gate gave for a held step on a web site. |
| confirm | boolean | – | add/remove a restriction: true = do it. Omit to preview. |
| days | – | – | add: only on these weekdays: [5,6], ["fri","sat"] or "sun-thu" (0 = Sunday). |
| effect | string | – | add: what happens when an agent touches it. |
| folder | string | – | add: same as path — the folder (or file) the rule is for. ~/Name = in the home folder; **/Name = any folder with that name. |
| hours | – | – | add: only at these hours on this computer's clock: [18,8] or "18-8" (wraps overnight). |
| level | string | – | level: the approval level to take — strict, routine or production. Omit to see the current one and what each does. |
| n | integer | – | remove: the rule number from 'list'. |
| note | string | – | add: a few words the user will see when the rule decides. |
| path | string | – | add: the file or folder (a folder covers everything inside it). |
| server | string | – | add: one MCP server whose named tools are trusted past the hold after outside content (allow only): its name as in mcp__<server>__<tool>, e.g. "dreamz" or "plugin:dreamz:dreamz". Needs tools. |
| set | string | – | set: the ready-made rule set to add ('developer'). |
| site | string | – | add: a web site instead of a folder: "example.com" (its subdomains too), "https://shop.example.com", or "*" for every site. Effects: allow, ask, deny. |
| step | string | – | add: a kind of outbound step instead of a folder or site: install, install_locked, push, publish, send, launch_agent, instructions, click, fetch, mcp_send. Effects: allow (answers only the hold after… |
| sub_agent | – | – | add: which sub-agent inside the agent: "main" = the main conversation, "all but main" = only sub-agents. |
| tools | – | – | add, with server: the exact tools to trust, ["render_take","take_status"]. No wildcard; a messaging tool (send, reply, email, post, share …) is refused. |
| when | array | – | add: conditions, all must hold (optional). |
No output schema declared.
No examples provided.
nodata_run Run a command with vault secrets ~491
Run a command on this machine (tests, a migration, a deploy script) WITH the project's secrets from the NoData vault as environment variables — without you seeing them. The secrets are pulled into memory only (no .env is written); you get back the exit code and the output with every secret value blanked. EVERY command needs the user's approval, as written. When a key's policy says a person approves each use, or the client cannot ask the user itself, NoData asks the person: a page opens in their browser (or give them the link this returns), they sign in, check the command and the code, and approve it there — you cannot approve it, and there is no code for you to send back. Then call nodata_run again with the same command, folder and project; it resumes the same request. Never embed a secret value in the command, never try to print or encode one, and never confirm on the user's behalf. Uses this computer's NoData connection (`npx -y @nodatachat/mcp connect`, approved once in the browser) — no separate token; a project-limited CLI token saved with `secrets-login` is used first if present. Never ask the user for a key in the chat.
| Name | Type | Req | Description |
|---|---|---|---|
| approval | string | – | Optional: the approval id NoData returned for this exact command. It releases nothing by itself — only a person's approval of this command at NoData does. |
| command | string | yes | The exact command line to run, e.g. `npm test`. |
| cost_usd | number | – | Optional: this run's price in dollars, as you showed it to the user. A key whose rule caps its daily spend counts it and refuses a run that states none. State the real price; it is kept on the receip… |
| cwd | string | – | Folder to run in. Default: the current folder. |
| partner_of | string | – | Optional: a CLIENT organisation's id — run with that client's keys, as its partner. The client's own key board decides each key (only a rule naming this computer releases one; its admins approve, nev… |
| project | string | – | Secrets project label. Default: the token's project (a connected computer has none: name it). |
| timeout_seconds | integer | – | Default 300. |
No output schema declared.
No examples provided.
nodata_scan_feedback Report a wrong scan result ~233
REPORT a wrong scan result so NoData's detectors improve: files the scan flagged that are NOT actually sensitive (`not_sensitive`: relative paths or patterns from the scan), and kinds of data it MISSED (`missed`: {"phone": 2}). Send only what the user said — never guess on their behalf. Content-free: only kind labels and counts leave this machine (e.g. "not_sensitive:ID×2 · missed:phone×1"), never a value, a file name or free text. It does not change any file or any plan; to leave a file alone use nodata_plan with "keep".
| Name | Type | Req | Description |
|---|---|---|---|
| missed | object | – | Kinds the scan did not catch, with how many the user saw: id, phone, email, card, iban, bank_account, name, address, birth_date, medical, plate, secret, amount, business_term, other. |
| not_sensitive | array | – | Relative paths (or patterns like "fixtures/**") whose findings are false alarms, as the scan listed them. |
| path | string | yes | The folder or file that was scanned. |
No output schema declared.
No examples provided.
nodata_scan_folder Scan a folder for sensitive data ~821
SEE what's sensitive. Scans a LOCAL folder for hardcoded secrets and unencrypted PII (checksum-validated Israeli ID, Luhn credit card, IBAN, phones), entirely on this machine. It sorts every file by what it is and reads it: Office (2007+ and 97-2003), OpenDocument, RTF, PDF, Outlook mail with its attachments, files inside ZIP archives, and pictures by OCR on this computer (Tesseract when installed, otherwise a built-in reader fetched once from npm — nothing to install by hand). Every file NOT read is counted and named with why (COVERAGE) — relay that to the user; never say everything was checked while a row says NOT. Returns the WHOLE result — counts by type, by folder and by file — and writes a full content-free HTML report (every finding) to ~/.nodata/reports, with a link to open it. A folder of more than 200 files is MAPPED first (parts, file counts, types) and scanned part by part in rounds: on CONTINUE call again with the same path, without asking, until ALL PARTS DONE; the last round gives the whole result. Ends with the IRON CHAIN block: the scan's QUALITY, a VERDICT computed from it (READY / NOT READY / NOT PROVEN / UNKNOWN) and the one mandatory NEXT step: relay it as it is. When the verdict is NOT READY the same answer carries the recommended policy and its PATH TO READY (STEP 5): show it and ask for approval. Do not list, count or glob the folder yourself before it: its first answer is the map. GUIDED: when the client can show dialogs, call it with NO path when the user did not name a folder: it asks the person which folder, and after a NOT READY verdict it asks the person what to do (a clean copy preselected, explained in one sentence) and runs that choice itself. Relay its answer; do not ask the same questions yourself. Every scan leaves a receipt (counts only). DEMO: when the user wants to see, show, try or demo NoData ("show me how it protects data from AI") and names no folder of their own, call it with demo:true: it writes a folder of invented data…
| Name | Type | Req | Description |
|---|---|---|---|
| demo | boolean | – | (Re)write the demo and trial folder of INVENTED data (~/NoData Demo and Trial: salaries, a customer letter, a medical summary, app secrets, one ordinary file) and scan it. Use it when the user wants… |
| git_history | boolean | – | Also check the folder's git history for secrets that were committed and later deleted (they stay in every clone). Content-free: rule, file, commit id. |
| include_tool_dirs | boolean | – | Also scan tool / cache folders (.cache, .cargo, editor extensions, AppData …). Default false — they hold samples, not the user's files. |
| max_files | integer | – | Files to walk before stopping. Default 5000; raise it for a home folder or a large drive. |
| path | string | – | Folder or single file to scan. Omit it when the user did not name one: a client with dialogs asks the person; otherwise the directory the server runs in. |
| restart | boolean | – | A folder being scanned in parts: start again from the first part. Default false (continue where the last round ended). |
No output schema declared.
No examples provided.
nodata_scan_terms Set the words that make a file sensitive ~533
The SCAN TERMS: words, phrases or symbols that make a file sensitive in nodata_scan_folder and nodata_plan, each with a label and a tier (1 public · 2 internal · 3 confidential · 4 restricted). The organisation's terms come from NoData when this computer is connected (cached here); the user's own are kept on this computer. A file holding a term is reported with the term's label, tier and count (never the text around it) and is raised to that tier in the plan. action 'search' answers "would the scan catch X?": how a column named X is judged, whether X is in the built-in list, in the user's or the organisation's terms, and (with path) how many files in a folder hold it, counts only; when nothing covers it, it gives the add call. Use it when the user names something sensitive the scan did not report. action 'amounts' with on:true treats money amounts as sensitive (the scan reports them, a clean copy replaces them with [AMOUNT]); off by default, because a clean copy is for working with the numbers. Turning it on asks for confirm:true; turning it off needs the user's own approval. Use it when the amounts ARE the secret (holdings, a family office). action 'list' shows both lists. 'add' adds one of the user's terms (asks for confirm:true). 'remove' removes one of the user's terms by id (needs the user's own approval). The organisation's terms are changed only at NoData.
| Name | Type | Req | Description |
|---|---|---|---|
| action | string | yes | – |
| approval | string | – | remove: the code the user approved. |
| confirm | boolean | – | add: true = do it. Omit to preview. |
| id | string | – | add: a short id (letters, digits, - _); default from the label. remove: the term's id. |
| label | string | – | add: what it means, shown in the scan instead of the word. |
| match | string | – | add: the literal word, phrase or symbol (case and Hebrew vowel marks are ignored). |
| on | boolean | – | amounts: true = treat money amounts as sensitive, false = keep them (the default). |
| path | string | – | search: a folder (or file) to count it in, counts only. |
| query | string | – | search: the word, phrase or field name the user thinks is sensitive. |
| tier | integer | – | add: 1 public · 2 internal · 3 confidential · 4 restricted. |
No output schema declared.
No examples provided.
nodata_terms Show or accept NoData's terms ~181
Show NoData's terms of use (current version) and, on the user's explicit approval, accept them. Data tools stay off until the current version is accepted. ASKS FIRST: without confirm:true it only shows the terms summary and link — show it to the user and call again with confirm:true only if they say they accept. Records operational identity only (organisation, agent, version, time, hash), never content. LANGUAGE: if the user has not chosen one, ask them — עברית or English — and pass it as `language`; it is kept for every later session and every agent on this machine.
| Name | Type | Req | Description |
|---|---|---|---|
| confirm | boolean | – | true = the user accepts these terms. Omit to show them. |
| language | string | – | The language the user chose for NoData and their agents: he (עברית) or en (English). |
No output schema declared.
No examples provided.
How do I install the NoData MCP server?
NoData runs locally as an npm package, launched with npx -y @nodatachat/mcp. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation.
npm · @nodatachat/mcp
claude mcp add daviderez4-nodata -- npx -y @nodatachat/mcp
What is the NoData MCP server?
NoData is an MCP server listed in the public MCP registry as io.github.daviderez4/nodata. Governed access to sensitive data, secrets and APIs for AI agents. A signed proof for each decision. This page covers its npm package (@nodatachat/mcp).
Is the NoData MCP server safe to use?
NoData scores 63 out of 100 on VerifyMCP. We found no known CVEs affecting it as of 21 November 2026. It declares no install or post-install scripts. 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 NoData MCP server expose?
NoData exposes 15 tools: nodata_terms, nodata_node, nodata_get_started, nodata_plan, nodata_scan_feedback, and 10 more. Their descriptions and schemas cost roughly 6,885 tokens of context every time the server is loaded.
Is the NoData MCP server still maintained?
NoData 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 NoData MCP server under?
NoData declares the FSL-1.1-ALv2 licence, which is not on the OSI-approved list. Read the terms before using it at work, and note this covers the source only, not the cost of any service it calls.
What did VerifyMCP test on the NoData MCP server?
VerifyMCP ran 15 checks on NoData, covering Malware scan, Known CVEs, Install scripts, Dependency health and 11 more. One item could not be completed, a gap on our side rather than a finding about NoData: Security disclosure. Endpoint Security and Transport & Reachability do not apply to an npm package, so they were not tested. The score does not measure everything, and the scoring documentation lists what it leaves out.