Honcho MCP
Honcho MCP gives an agent memory about a user: it stores conclusions from conversations and answers questions about them.
- 8,102 ★ on the repository
- Last commit
- AGPL-3.0
Extended access
Why this level
- Conversation content and personal preferences are stored in an external memory service.
- The API key opens the whole workspace memory.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/honchoDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Conversation content and personal preferences are stored in an external memory service.
- The API key opens the whole workspace memory.
This is third-party code. Review the repository files before installing.
What it does
Honcho keeps a model of the people and agents in a conversation and answers questions about them. The hosted MCP server `https://mcp.honcho.dev` connects this memory to any MCP client: the assistant remembers user preferences and answers better from one conversation to the next. The server works in two modes. In recall mode the assistant only reads with `chat`, `workspace_chat`, `search`, `get_representation`, `get_peer_context` and `list_conclusions`. In memory-store mode the conversation itself becomes learning material and `create_session`, `create_peer`, `add_peers_to_session` and `add_messages_to_session` join the reading tools. The hosted server needs an API key from `app.honcho.dev` starting with `hch-`. You can also run the server yourself over stdio, HTTP or Docker.
Who it is for. People who want Claude, Cursor or Codex to remember preferences and context between sessions and are ready to keep that memory in Honcho.
Good fit when
- The assistant must remember how the user likes code formatted without reminders in every session.
- An app asks the memory what is known about a user and builds the answer around it.
- A conversation is recorded into a Honcho session so background reasoning enriches the picture of the person.
- The same memory is connected at once in Claude Desktop, Cursor and Windsurf.
Not a fit when
- User data cannot live in an external service: in memory-store mode conversations go to Honcho.
- Memory without a key or sign-up is needed: the hosted server needs a key from `app.honcho.dev`.
- A notes file in the repository is enough: a separate memory is overkill for a small project.
Example request
What do you know about me and my working style? Take it into account and propose a structure for the new module.MCP
- Transport
- http, stdio
- Authentication
- API key
Environment variables
HONCHO_API_KEYrequired, secret- Honcho API key starting with `hch-`, issued at app.honcho.dev
HONCHO_WORKSPACE_ID- Default workspace for the tools
Limitations
The hosted server needs an API key from `app.honcho.dev`, sent in the `Authorization: Bearer hch-...` header. Memory-store mode records both sides of a conversation, so personal data goes to Honcho, and if recording is unwanted stay in recall mode. Tools work inside a `workspace_id` workspace that can be set with the `X-Honcho-Workspace-ID` header. For Codex and Claude Desktop the hosted server connects through the `mcp-remote` bridge, started with `npx`. The visible effect of background reasoning appears after a few conversations, not at once.
How to disable. Remove the `honcho` entry from the client settings and revoke the key at `app.honcho.dev`. Accumulated data lives in Honcho and is deleted separately.
FAQ
How do I connect Claude Code?
Run `claude mcp add honcho --transport http "https://mcp.honcho.dev" --header "Authorization: Bearer hch-your-key" --header "X-Honcho-Plugin: claude-code"`. Get the key at `app.honcho.dev`.
What are recall and memory-store modes?
Recall only reads memory. Memory-store also creates sessions and records messages and is used when a conversation must be remembered.
Can I run the server myself?
Yes: over stdio, HTTP or Docker. Stdio needs Bun and runs as `bun --cwd /path/to/honcho/mcp src/stdio.ts` with `HONCHO_API_KEY`, `HONCHO_API_URL` and `HONCHO_WORKSPACE_ID`.
Why the `X-Honcho-Plugin` header?
It names the client that configured the connection. The header is optional, but bridges like `mcp-remote` send only the HTTP library name.
How do I verify the connection?
Ask the assistant what it knows about you. After a few conversations the answer becomes richer.
Discussion
No questions or reviews yet.
Related
MCP serversHoncho plugins from Plastic Labs: Claude Code memory that survives sessions and skills for wiring the Honcho SDK into your app.
The Memanto memory MCP server: an agent stores facts and decisions and finds them in later sessions, even as of a date.
Honcho memory plugin for OpenClaw: separate models of the user and the agent and honcho tools beside the built-in memory.
A self-hosted MCP memory server that keeps preferences and decisions across conversations in Claude Code, Cursor and other clients
An MCP server for long-term agent memory: compresses dialogues into atomic facts, resolves pronouns and anchors relative time to absolute dates
A local AI agent platform: agents, memory and MCP servers live in the project's .hyperchat folder, usable from the command line or the browser
README in short
The integration page covers the hosted address, setup for ten clients, local running, workspaces and a connection check.