Logseq MCP Server

A Model Context Protocol server that gives an

LLM configurable, safety-scoped access to a Logseq

graph: query and search broadly (all output filtered by a blacklist), but write

only inside the agent's own namespace plus a narrow task-status channel.

Built on FastMCP (the high-level API of the official mcp package).

Targets the file/Markdown ("OG") version of Logseq. The newer DB (SQLite)

version changed the underlying schema; some methods may behave differently

there.

Requirements

• A running Logseq with the local HTTP API server enabled

(Settings → Features → *HTTP APIs server*, then start it from the 🔌 menu).

• An authorization token created in the HTTP API server settings.

Usage with Claude Desktop

{
  "mcpServers": {
    "logseq": {
      "command": "uvx",
      "args": ["mcp-server-logseq"],
      "env": {
        "LOGSEQ_API_TOKEN": "",
        "LOGSEQ_API_URL": "http://127.0.0.1:12315"
      }
    }
  }
}

Configuration

The token is read from the environment or --api-key; it is never stored in

code. A .env file is supported (see .env.example).

Config file (optional)

Behaviour beyond the defaults is set in a TOML file — path from

LOGSEQ_MCP_CONFIG (default ~/.config/logseq-mcp/config.toml). Custom queries

live in EDN files next to it. The server runs fine with no config file (safe

read-mostly defaults); see [examples/config.toml](examples/config.toml) for a

full annotated example.

Secrets and the API URL stay in the environment, never in this file.

Transports

By default the server runs over stdio (for Claude Desktop and other local

clients). A Streamable HTTP transport is also available for remote/networked

use (e.g. a phone client):

LOGSEQ_MCP_HTTP_TOKEN= \
  mcp-server-logseq --transport streamable-http --host 0.0.0.0 --port 8000
# MCP endpoint: http://:8000/mcp

Env vars: LOGSEQ_MCP_TRANSPORT, LOGSEQ_MCP_HOST, LOGSEQ_MCP_PORT,

LOGSEQ_MCP_HTTP_TOKEN (or --http-token).

Authentication

The Streamable HTTP transport requires a bearer token: every request must

send Authorization: Bearer , or it gets 401. The

server refuses to start in this mode without a token set. Note this is a

distinct secret from LOGSEQ_API_TOKEN:

⚠️ A bearer token over plain HTTP is only safe on an already-encrypted

channel. Don't expose the raw port to the open internet. The easy path for a

home/headless host is Tailscale: install it on the host and the client,

and reach http://..ts.net:8000/mcp over the encrypted

tunnel — no domains, nginx, or certificates. (tailscale serve can add TLS

if you want https://.)

Docker

Build once:

docker build -t logseq-mcp .

Quick try (ephemeral — --rm removes the container on stop):

docker run --rm -p 8000:8000 \
  -e LOGSEQ_API_TOKEN= \
  -e LOGSEQ_MCP_HTTP_TOKEN= \
  -e TZ=Europe/Moscow \
  logseq-mcp

Persistent deploy (e.g. a headless Mac mini) — run once; --restart brings it

back after reboots:

docker run -d --name logseq-mcp --restart unless-stopped -p 8000:8000 \
  -e LOGSEQ_API_TOKEN= \
  -e LOGSEQ_MCP_HTTP_TOKEN= \
  -e TZ=Europe/Moscow \
  -e LOGSEQ_MCP_CONFIG=/cfg/config.toml \
  -v /path/to/config-dir:/cfg:ro \
  -v "/path/to/your/graph:/graph:ro" \
  logseq-mcp

• -v .../config-dir:/cfg — folder holding your config.toml (+ queries/,

rules/); set files_path = "/graph" in it to enable file search. Omit both

the mount and LOGSEQ_MCP_CONFIG to run on defaults.

• -v .../graph:/graph — your Logseq graph folder (read-only), for file search.
• -e TZ= — local time for audit-log timestamps (image bundles tzdata;

the clock is UTC otherwise).

The container serves Streamable HTTP on port 8000 and talks to a Logseq running

on the host. On Docker Desktop (macOS/Windows) the default

LOGSEQ_API_URL=http://host.docker.internal:12315 already points at the host;

on Linux add --add-host=host.docker.internal:host-gateway (or set

LOGSEQ_API_URL to the host IP). Make sure Logseq's HTTP API server is running

and listening.

Tools

All read output is normalized to a flat JSON shape and passed through the

blacklist. Reads resolve ((block refs)) non-lossily (the resolved block's

uuid/status is kept so you can act on it).

Find

• search — full-text search over block content (query, regex?, limit?,

case_sensitive?, exclude_journals?). Uses ripgrep over files_path when

set, else a datascript content match.

• find_tasks — task blocks by markers?, tag?, under_tag? (descendant),

page?, priority?, limit?.

• list_pages — page names under a namespace prefix? (depth? limits levels).

Discovers a namespace's child pages, which are separate pages a parent's

read_page won't show. Structure only, not block content.

• custom_query — run a named query from the config (name, inputs?).
• list_custom_queries — list the configured queries.
• datascript_query — run a raw Datalog query (query, inputs?, rules?).

Guide

• get_logseq_guide — returns the authoritative guide for querying/writing this

graph (verified Datalog gotchas: lowercase names, prefix descendants, marker and

journal-day types, tags vs refs, read/write scoping). A single source of truth

co-located with the server, so agents don't re-derive (and mis-derive) behaviour.

Read

• read_page — a page as a normalized block tree (page, depth?).
• read_block — a block and its children (uuid, depth?).

Write (agent namespace only)

• write_note — create/append/replace a page under agent_write_prefix

(subpath, content?, mode?, properties?).

• set_page_properties — set/remove page properties (subpath, properties;

a null value removes one).

Tasks

• set_task_status — change only a task's marker (uuid, status); gated by

[tasks].allow_status_change.

Dynamic

• query_<name> — each config query with register_as_tool = true is

exposed as its own tool.

Development

git clone https://github.com/dailydaniel/logseq-mcp.git
cd logseq-mcp
cp .env.example .env   # fill in LOGSEQ_API_TOKEN
uv sync
uv run mcp-server-logseq

Inspect with the MCP Inspector:

npx @modelcontextprotocol/inspector uv --directory . run mcp-server-logseq

License

MIT