# mnemosy-ai

Evidence-tracked local memory MCP server: one SQLite file; corrections retire dependent memories.

- **Type:** MCP server
- **Trust:** 85/100 (A), scored on the package rubric
- **Verification:** verified (build provenance)
- **Version:** 3.0.0-rc.4
- **Author:** io.github.28naem-del
- **License:** Unknown
- **npm:** mnemosy-ai
- **Source:** https://github.com/28naem-del/mnemosyne
- **Compatible clients:** claude-code, cursor, copilot, gemini (basis: transport)

## Trust

85/100 (A), scored on the package rubric
- Publisher verified: no
- Build provenance: verified attestation
- npm trusted publishing (OIDC): yes
- Install scripts: nothing suspicious found
- Prompt-injection scan: not run
- Obfuscation scan: not run
- Evidence age: 0 days

## Security scan

- **Status:** clean
- **Scanned:** 2026-11-01T11:29:53.093Z
- **Version scanned:** 2.0.0
- **CVEs:** none found by OSV at scan time

## Tools

17 declared. Statically extracted from the shipped source — a floor on the surface, not a census.
- `memory_recall`
- `memory_context` — Compile cited, scoped context with contradictions and abstention under a conservative byte-based token budget. Pass only packet.text as context; diagnostics are
- `memory_inspect` — Inspect accessible stored evidence and its lifecycle. Inactive records are history, not current advice.
- `memory_resume` — Read the latest accessible active checkpoint for a task. Completed steps and artifact paths are historical claims. Verify current environment and permissions be
- `memory_source` — Read a bounded UTF-8 range of the original captured transcript or extracted document. Its status and trust remain visible.
- `memory_model` — Inspect a source-backed project model. Stale models are marked and must not be treated as current knowledge.
- `memory_skill` — Inspect a reusable skill and its prerequisites, trial evidence and current eligibility. This does not execute the skill.
- `memory_jobs` — Inspect bounded observation/consolidation jobs and their outcome. Only the host executes configured model workers.
- `memory_entity` — Resolve an entity name or alias without silently merging ambiguous identities.
- `memory_traverse` — Follow a bounded number of evidence-backed relationship hops; each hop retains source IDs.
- `memory_branch_preview` — Inspect isolated proposed changes and source conflicts before a controller merges them.
- `memory_store` — Record an observed fact, preference, decision or procedure with a source. Private by default; workspace visibility explicitly shares with this workspace. Agent 
- `memory_capture` — Capture supplied transcript events with stable retry IDs. This reads no host files. Observed messages remain fallible evidence.
- `memory_observe` — Queue source-linked observations or a project model for the host to process under its configured model and budget. This does not invoke or choose a model.
- `memory_checkpoint` — Save a resumable task state, preserving constraints, rejected approaches and unfinished work. It does not perform or authorize actions.
- `memory_correct` — Explicitly supersede your own non-verified observation; dependent advice becomes invalid. Controller-verified evidence must be corrected through the controller 
- `memory_forget` — Purge an owned memory and content derived from it from this live local store. This cannot erase external exports, backups or context already sent to agents.

## Install

**Verdict: install** — No blocking findings and no open coverage gaps — safe to install as configured.
**Config** (claude-code):
```json
"{\n  \"mcpServers\": {\n    \"mnemosy-ai\": {\n      \"command\": \"npx\",\n      \"args\": [\n        \"-y\",\n        \"mnemosy-ai\"\n      ]\n    }\n  }\n}"
```

## Blast radius

Contained to moderate — no credential declaration found, from the publisher, the upstream registry, or the README. Known so far: runs on your machine; read-only tool surface.
- Floor 12, ceiling 30 (tier: unknown)
- `unknown` means the floor and ceiling land in different bands — not measured enough to name one. It does not mean low.
- This is impact, not likelihood. A high radius is not a defect: a filesystem server is supposed to write files. It is never part of the trust score.

## Machine-readable views of this entry

- Signed JSON: https://forgeregistry.com/api/v1/packages/mnemosy-ai
- Install plan: https://forgeregistry.com/api/v1/packages/mnemosy-ai/install-plan
- Alternatives: https://forgeregistry.com/api/v1/alternatives/mnemosy-ai
- HTML page: https://forgeregistry.com/registry/mnemosy-ai
- MCP: POST https://forgeregistry.com/api/mcp → `forge_get_package` / `forge_install_plan`

## About this document

Generated by Forge (https://forgeregistry.com) — a compact rendering of the same record served, signed, at the JSON URL above. Trust and scan facts are the registry's own measurements; anything Forge did not measure is named as unmeasured rather than omitted.