Logseq MCP Server

What is Logseq MCP Server?

A Model Context Protocol server that gives a language model configurable, safety-scoped access to a Logseq graph—querying and searching broadly with output filtered by a blacklist, but writing only inside the agent’s own namespace plus a narrow task-status channel. It targets the file/Markdown version of Logseq.

How to use Logseq MCP Server?

Install with uvx mcp-server-logseq and configure via environment variables (LOGSEQ_API_TOKEN, LOGSEQ_API_URL) or CLI flags. Requires a running Logseq with the local HTTP API server enabled and an authorization token. Runs over stdio by default; also supports Streamable HTTP for remote use.

Key features of Logseq MCP Server

• Full-text search, task lookup, and page listing with blacklist filtering
• Read pages, blocks, and run custom Datalog queries
• Write notes only in the agent’s own namespace prefix
• Gateable task status changes
• Dynamic tools generated from config queries
• Optional audit log recorded to today’s journal
• Supports both stdio and Streamable HTTP transports

Use cases of Logseq MCP Server

• Letting an AI assistant safely search and browse a Logseq knowledge base
• Allowing an agent to write notes or set page properties in its own namespace
• Enabling an agent to update task markers (e.g., TODO → DONE) under configurable control
• Running predefined or custom Datalog queries against the graph

FAQ from Logseq MCP Server

What version of Logseq does the server target?

It targets the file/Markdown (“OG”) version of Logseq. The newer DB (SQLite) version changed the schema, so some methods may behave differently.

What are the runtime requirements?

A running Logseq with its HTTP API server enabled, an authorization token, and either uvx or Docker. No config file is needed for defaults.

What transports and authentication are available?

Stdio (default, no extra auth). Streamable HTTP requires a bearer token (LOGSEQ_MCP_HTTP_TOKEN) distinct from the Logseq API token; the server refuses to start without it. Streamable HTTP over plain HTTP should only be used on an already-encrypted channel (e.g., a Tailscale tunnel).

Where are secrets and configuration stored?

Secrets (LOGSEQ_API_TOKEN, LOGSEQ_MCP_HTTP_TOKEN) are set in the environment or CLI, never in code or the config file. An optional TOML file at ~/.config/logseq-mcp/config.toml controls behaviour beyond defaults.

What are the known limits?

All write operations are scoped to the agent namespace prefix (default byAgent) except for the gated set_task_status tool. Reads are filtered by a configurable page blacklist.