# IHMT Memory (pypi · ihmt-memory) Long-term memory for AI coding agents, stored as a tree of plain local files. - Trust score: 53/100 (low) - Registry status: active - Liveness: live - Owner verified: no - Last scored: 2026-10-17 ## Components - pypi · `ihmt-memory`: 53/100 (this document), [markdown](https://verifymcp.io/servers/gonzaroman-ihmt-memory/ihmt-memory.md), [page](https://verifymcp.io/servers/gonzaroman-ihmt-memory/ihmt-memory) ## Channel facts - Registry: `pypi` - Package: `ihmt-memory` - Version: `1.2.1` - Transport: `stdio` ## Trust breakdown 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. Scores are 0–100 per category. Scoring method: https://verifymcp.io/docs/scoring (what has changed: https://verifymcp.io/docs/scoring/changelog) Scored 2026-10-17. - **Supply Chain Security**: 50/100 - Malware scan not yet available for this package. - No known CVEs affecting this package version or its production dependencies. - Runs hatchling.build at install time, a recognised build step with no custom scripting around it. - 1 of 28 dependencies flagged as unhealthy. - **Provenance & Transparency**: 45/100 - Source repository is publicly reachable at the declared URL. - Provenance check failed: no build-provenance attestation is published. - Clear OSI-approved license (MIT). - Actively maintained (last published 1 days ago). - Disclosure check failed: no security disclosure policy was found in the source repository. - **Schema Quality & AI Usability**: 64/100 - AI-judged instruction clarity (good). - Context-footprint check failed: tool/resource definitions use about 1505 tokens (~167/item across 9 items; 9 tools + 0 resources), over budget; trim descriptions and params. - Usage-examples check failed: none of the tools include examples. - **Stability & Change Management**: 10/100 - Stability observed for 3 of 30 days with no destabilising changes; credit accrues until the full window elapses. - **Tool Coverage**: 71/100 - 100% of tools have a non-trivial description (not blank, and not just the tool's name). - 0% of tool parameters carry a description. - Structured output schemas are declared (100% of tools); any adoption earns full credit. - **Tool Safety**: 100/100 - No prompt-injection markers were found in the server instructions, tool names or descriptions we captured. - We read all 9 captured tool definition(s), and no name or description among them implies an irreversible operation. - An AI judge read all 10 captured unit(s) of tool text and found none that tries to manipulate the model reading it. - **Capabilities**: 100/100 - Implements a current MCP spec version (2026-07-28). ## Install ### How do I install the IHMT Memory MCP server? IHMT Memory runs locally as a PyPI package, launched with uvx ihmt-memory. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation. ### Claude ```bash claude mcp add gonzaroman-ihmt-memory -- uvx ihmt-memory ``` ### Cursor ```json { "mcpServers": { "gonzaroman-ihmt-memory": { "command": "uvx", "args": [ "ihmt-memory" ] } } } ``` ### VS Code ```json { "servers": { "gonzaroman-ihmt-memory": { "command": "uvx", "args": [ "ihmt-memory" ] } } } ``` ### Codex ```bash codex mcp add gonzaroman-ihmt-memory -- uvx ihmt-memory ``` ### opencode ```json { "$schema": "https://opencode.ai/config.json", "mcp": { "gonzaroman-ihmt-memory": { "type": "local", "command": [ "uvx", "ihmt-memory" ], "enabled": true } } } ``` ### OpenClaw ```bash openclaw mcp add gonzaroman-ihmt-memory --command uvx --arg ihmt-memory ``` ### Hermes ```yaml mcp_servers: gonzaroman-ihmt-memory: command: "uvx" args: ["ihmt-memory"] ``` ### Netclaw ```json { "McpServers": { "gonzaroman-ihmt-memory": { "Transport": "stdio", "Command": "uvx", "Arguments": [ "ihmt-memory" ] } } } ``` ### Vellum ```bash assistant mcp add gonzaroman-ihmt-memory -t stdio -c uvx -a ihmt-memory ``` ### Other ```json { "mcpServers": { "gonzaroman-ihmt-memory": { "command": "uvx", "args": [ "ihmt-memory" ] } } } ``` ## Changelog Every change recorded for this component, newest first. Days that predate change tracking, or that we cannot explain, say so: "we were watching and nothing happened" and "we were not watching" are different claims. ### 2026-10-17 (score 53, +1) No change was recorded against any check on this day. Stability & Change Management went from 7 to 10. That category is still filling its 30-day observation window: 2 days of observed history at the previous scan, 3 at this one. The score rises as the window fills, whether or not the server changes. ### 2026-10-16 (score 52, −15) - [security regression] Malware scan: pass → unverified ### 2026-10-15 (score 67, +16) - [security regression] Tool safety: pass → unverified - [security improvement] Malware scan: unverified → pass - [security] Stability: Stability not yet verified: we do not have a sandbox capture of the MCP schema this version of the package serves yet. - [functional regression] Capabilities: pass → unverified - [functional regression] Tool coverage: 100 → unverified - [functional improvement] Stability: unverified → 0.03 - [functional] Package version: 1.2.0 → 1.2.1 ### 2026-10-14 (score 51, +27) - [security improvement] Injection markers: unverified → pass - [security] First check of Judged manipulation: pass - [security] Stability: Stability not yet verified: not enough scan history yet (needs a 30-day window). - [functional improvement] Tool coverage: unverified → 100 - [functional improvement] MCP protocol: unverified → pass - [functional] First check of Tool coverage: 0 - [functional] First check of Schema quality: fail - [functional] First check of Destructive annotations: pass - [functional] First check of Schema quality: fail - [functional] First check of Schema quality: good - [functional] First check of Tool coverage: 100 ### 2026-10-13 (score 24) First indexed and scored. ## MCP tools (9) ### `search_memory` (~163 tokens) Search the user's long-term memory (IHMT): decisions, setup, preferences, people, earlier sessions. Use several distinctive words, not a bare name. Returns the best memory with its date (plus one line per other match). AMBIGUOUS means several memories fit equally: ask the user or call again with `clue` (a place, date, project). Always relay OUTDATED notices. Args: query: What to look for, in any language. clue: Distinguishing detail, only after an AMBIGUOUS reply. detail: "compact" (default) or "full" (adds ids, tree paths and excerpts). Input parameters: - `clue` - `detail` (string) - `query` (string, required) Output parameters: - `result` (string) ### `save_memory` (~257 tokens) Store durable knowledge in the user's long-term memory (IHMT). Save decisions and why, setup, preferences, people and projects, procedures. Not transient chat, not secrets. Write it to make sense months later, and start with the date ("2026-02-03: ...") when it belongs to one. Reports how it was filed and any contradiction with an earlier memory — relay that notice. Args: content: The text to remember, self-contained. domain: Leave "general" to let IHMT classify it, or e.g. "software.java". content_type: "auto" (recommended), "code", "narrative", "clinical", "personal", "process" or "generic". replaces: When this corrects or updates something saved earlier, a few distinctive words describing that earlier memory ("RainDesktop version 1.3"). It is then flagged OUTDATED and searches answer with this one first. If several memories fit, nothing is flagged and the candidates are listed for `mark_outdated`. Input parameters: - `content` (string, required) - `content_type` (string) - `domain` (string) - `replaces` (string) Output parameters: - `result` (string) ### `mark_outdated` (~109 tokens) Flag one memory as corrected by another, after save_memory listed candidates. The old memory is kept (it was true on its date) but carries an OUTDATED notice, and searches answer with the new one first. Use the ids save_memory or a full search printed. Args: old_id: Id of the memory that is no longer current. new_id: Id of the memory that replaces it. Input parameters: - `new_id` (string, required) - `old_id` (string, required) Output parameters: - `result` (string) ### `project_map` (~137 tokens) Compact map of a codebase: files with their size in tokens, optionally their symbols and line ranges. Call it before exploring a project instead of listing directories and opening files. Indexes the project on first use and resyncs changed files by checksum on every call. Args: path: Project root directory (absolute). detail: "files" (cheapest) or "symbols" (adds methods/classes with first-last line). subpath: Restrict to a directory inside the project, e.g. "src/main". Input parameters: - `detail` (string) - `path` (string, required) - `subpath` (string) Output parameters: - `result` (string) ### `find_code` (~220 tokens) Find the method or class that answers a question about a codebase; returns only that symbol. Output: "file:first-last Symbol" and its code, instead of whole files. Name the class or method if you know it ("PedidoJpaMapper aDominio"). AMBIGUOUS lists candidates in different files: refine the query or pass `clue`. The index resyncs by checksum, so code is never stale. Args: query: What you are looking for: identifiers plus a few words. path: Project root directory (absolute). scope: "main" (skip tests, default), "test" or "all". subpath: Restrict to a directory inside the project. clue: Extra detail after an AMBIGUOUS reply. max_tokens: Cap on the returned code. Input parameters: - `clue` - `max_tokens` (integer) - `path` (string, required) - `query` (string, required) - `scope` (string) - `subpath` (string) Output parameters: - `result` (string) ### `read_file` (~199 tokens) Read a text file, remembering what you were given this session. Reading it again returns "UNCHANGED" (a few tokens) or only a unified diff against your earlier copy. Prefer it over re-reading a file you already read in this session. Pass force=true if the earlier copy is no longer in your context (e.g. after a context compaction). With start_line/end_line (e.g. a range from project_map) only those lines are returned. Args: path: Absolute path of the file. force: Return the full content even if it has not changed. start_line: First line to return (1-based); 0 for the whole file. end_line: Last line to return (inclusive); 0 for the end of the file. Input parameters: - `end_line` (integer) - `force` (boolean) - `path` (string, required) - `start_line` (integer) Output parameters: - `result` (string) ### `note` (~72 tokens) Keep something you learned in this session (where a thing lives, what a method does, a result). Session notes are cheap to recall later and survive a context compaction; they vanish when the session ends. Include file:line when relevant. Durable facts belong in save_memory instead. Input parameters: - `text` (string, required) Output parameters: - `result` (string) ### `recall` (~56 tokens) Search what you kept earlier in this session with note or digest_output (not long-term memory). Cheaper than re-reading files or re-running commands to remember what you already found out. Input parameters: - `clue` - `query` (string, required) Output parameters: - `result` (string) ### `digest_output` (~110 tokens) Condense a long command output (build log, test report) and keep the full text for recall. Returns the first lines, every line that mentions errors, failures, warnings or test totals, and the last lines. The whole output stays queryable with recall("