Best for
- Stateful agents that need cross-session memory
- Applications requiring personalization and long-term context
- Developers using MCP-compatible clients
- Teams choosing between a managed or self-hosted memory service
H
55FMRS · C
This official Honcho MCP server provides stateful-agent memory over streamable HTTP. The source documents the remote endpoint and authentication requirements, but does not list MCP tool identifiers, so the tool list is empty.
Honcho is memory infrastructure for building stateful agents. It stores messages, events, documents, or tool traces, reasons over them in the background, and builds evolving representations of people, agents, groups, projects, and ideas. Applications can query context, search results, representations, or natural-language insights. It is available as a managed service or as a self-hosted FastAPI server.
Tools
The tool list has not been reviewed yet.
Setup
Connect an MCP-compatible client to https://mcp.honcho.dev. The Authorization header is required and must use Bearer <token>; the token may be an OAuth access token or a Honcho API key from app.honcho.dev, whose keys start with hch-. Optionally set X-Honcho-Workspace-ID to provide a default workspace.
claude_desktop_config.json
{"mcpServers":{"honcho":{"url":"https://mcp.honcho.dev","headers":{"Authorization":"Bearer <token>"}}}}
Fit and risk
Not for
- One-off conversational applications that do not retain history
- Integration reviews requiring a documented list of MCP tool names and parameters
- Workloads where messages or documents must not be sent to an external memory service
Required permissions
- Requires an Authorization request header to mcp.honcho.dev
- Requires a valid OAuth access token or Honcho API key
- May optionally require a default workspace ID
- The service may receive and process session messages, events, documents, or tool traces
Risks and side effects
- Content sent to Honcho may contain sensitive personal or application data
- Background reasoning is asynchronous, so newly added messages may not immediately appear in chat or representation results
- The managed service depends on network access, credentials, and Honcho availability
- The AGPL-3.0 license may affect self-hosting and redistribution plans
Troubleshooting
- Verify that Authorization uses the Bearer <token> format
- Confirm that the token is valid and that Honcho API keys start with hch-
- Confirm that the MCP client supports streamable HTTP
- Check whether X-Honcho-Workspace-ID points to the intended workspace
- After writing new messages, allow time for background queue processing
- For self-hosting, check the database, required LLM API keys, migrations, API server, and background deriver
Use cases
Give coding agents persistent memory
Store conversations, events, documents, and tool traces
Query long-term representations and insights about users or agents
Generate session context for injection into LLM prompts
Perform hybrid message search
Supported clients
Claude CodeFull support
CursorPartial support
ClinePartial support
WindsurfPartial support
OpenCodePartial support
OpenClawPartial support
HermesPartial support