Agent Memory MCP
Local SQLite memory and project-docs search for coding agents via MCP
- Pricing
- Open source
- Platforms
- CLI, macOS, Linux
- Version
- 0.18.0
- Listing updated
- today
Similar MCP servers
About Agent Memory MCP
Agent Memory MCP is an open-source MCP server for knowledge and memory. It is available on CLI, macOS, Linux, Windows, vscode, self-hosted and API.
The README says this server gives coding agents a long-term memory of decisions, facts and procedures, and searches it together with project documents such as ADRs and runbooks. Each hit carries a trust score showing how far it can be relied on, so an agent can tell a recorded decision from a loose note.
It runs as one local binary backed by SQLite. Claude Code users can install it as a plugin with a local database per project; other agents such as Codex, Cursor, Gemini CLI and VS Code connect through a standalone setup that needs one embedding provider. An optional HTTP mode is available for a shared team instance.
Maintenance tools cover the memory's upkeep. A stewardship run looks for duplicates, contradictions and stale entries and routes risky changes to a review inbox, while a drift scan flags memories whose source file has changed or gone. Notes only become canonical through an explicit promotion step, so an agent cannot verify its own notes.
Tools this server provides (3)
Taken from the project documentation; the running server may differ.
- steward_run
- Finds duplicates, contradictions and stale entries in memory.
- drift_scan
- Flags memories whose source file changed or disappeared.
- promote_to_canonical
- Promotes a note to canonical status.
What it does
- Combined memory and docs search
- Ranks stored decisions, facts and procedures alongside project docs, ADRs and runbooks.
- Trust scoring on results
- Each search hit reports its layer, source, confidence and freshness.
- Memory stewardship runs
- The steward_run tool finds duplicates, contradictions and stale entries and queues risky items for review.
- Source drift detection
- The drift_scan tool flags memories whose source file changed or disappeared.
- Manual promotion to canonical
- Notes become canonical only through promote_to_canonical, so agents cannot self-verify.
- Local SQLite storage
- Runs as a single binary with a SQLite database, with an optional HTTP mode for teams.
Good for
- Recalling why an architectural or deployment decision was made
- Finding the runbook that matches a live incident
- Detecting when agent memory no longer matches the repository
- Sharing project memory across a team through a hosted HTTP instance
Pricing
Open source
MIT licensed; free to run locally.
Agent Memory MCP FAQ
Does it need a cloud service?
The README says it stays local, using one binary and SQLite. A standalone install does need one embedding provider, which the README does not name further.
Which agents are supported?
The README lists Codex, Cursor, Gemini CLI, OpenCode, VS Code, Windsurf and Claude Code, with a dedicated plugin for Claude Code.
Can an agent mark its own notes as verified?
No. The README says notes become canonical only through promote_to_canonical and that an agent cannot verify its own notes.
Is Agent Memory MCP free?
Yes. Agent Memory MCP is open source, so you can use and self-host it for free.
What are the best alternatives to Agent Memory MCP?
The closest alternatives to Agent Memory MCP are MCP Memory, Cross-Agent Memory, Total Agent Memory, mcp-memory-rs and Memoria — Deterministic AI Memory & Code Intelligence MCP.
Where this comes from
- Description: written from the project's README and its MCP Registry entry.