recall — Claude Code MCP server: what it does, how to install
Local persistent memory for AI agents without external API keys or cloud tokens. It maintains project history and manages multi-agent file locks.
claude mcp add recall -- uvx ai-recallworksWhat recall does
Recall provides offline persistent context by saving entries directly inside ~/.recall/. The server runs local ONNX embeddings on your machine, so semantic search and file retrieval need zero network traffic. Functions such as index_file and recall let coding assistants pull context from previous sessions alongside exact text citations.
The system requires Python 3.11+. The package exposes 22 MCP tools in total: 16 manage memories and project checkpoints.
The remaining six tools coordinate parallel agents. Functions like claim assign advisory locks with automatic TTL expiration to stop concurrent agents from overwriting the same script, while handoff transfers task intent across windows.
What it helps with
- retain architecture decisions across separate sessions
- index source files for local semantic retrieval with citations
- prevent file edit collisions when executing multiple agents
- transfer active context between planner and executor roles
Example prompts for the agent
- Find what we decided regarding database migrations last week.
- Index the documentation directory and check our rate limit policy.
- Claim src/auth.py so other running agents avoid modifying it during the refactor.
How to install recall
Platform CLI
claude mcp add recall -- uvx ai-recallworksuvx runs the command named after the package; if the README names another one, use uvx --from ai-recallworks <command>
Platform CLI
codex mcp add recall -- uvx ai-recallworksuvx runs the command named after the package; if the README names another one, use uvx --from ai-recallworks <command>
Config file
{
"mcpServers": {
"recall": {
"command": "uvx",
"args": [
"ai-recallworks"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from ai-recallworks <command>
Config file
{
"servers": {
"recall": {
"type": "stdio",
"command": "uvx",
"args": [
"ai-recallworks"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from ai-recallworks <command>
Platform CLI
gemini mcp add recall uvx ai-recallworksuvx runs the command named after the package; if the README names another one, use uvx --from ai-recallworks <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about recall
How to install recall in Claude Code?
Run claude mcp add recall -- uvx ai-recallworks in your terminal. It sets up the server via uvx without requiring external accounts.
Where does recall keep stored data?
Everything lives in your local ~/.recall/ folder. Vector embeddings are generated on-device with ONNX, so queries remain offline.
What do the multi-agent coordination tools do?
Commands like claim and release create time-limited advisory locks on files. They stop parallel AI assistants from overwriting identical files during complex tasks.
