Creativity · MCP — server

MCP Roam Research Server

The MCP Roam Research Server uses the Roam backend API (or the third-party Roam HTTP API) to query pages, create blocks, follow backlinks, and run Datalog queries. For Roam power users, this enables conversational knowledge capture: ask Claude to extract outlines from a meeting and append them as nested blocks under today's daily note.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: TypeScript
Transports: stdio

Capabilities

Tools: query_datalog, get_page, create_block, update_block, list_backlinks
Resources: roam://page/{title}, roam://block/{uid}
Auth: Roam backend API token + graph name

Install

npx -y @community/mcp-server-roam-research

Configuration

{
  "mcpServers": {
    "roam": {
      "command": "npx",
      "args": ["-y", "@community/mcp-server-roam-research"],
      "env": {
        "ROAM_GRAPH": "my-second-brain",
        "ROAM_TOKEN": "${ROAM_TOKEN}"
      }
    }
  }
}

Frequently asked questions

Why Datalog?

Roam stores blocks in a Datalog-like graph database. Exposing query_datalog gives the LLM expressive access (`find all blocks tagged #research under page X`) beyond what a flat REST API offers.

Is it safe to give write access?

Block writes in Roam are cheap and undoable, so write access is generally acceptable. Still, prefer creating under a specific parent to avoid polluting the daily note unintentionally.

How does it compare to Logseq or Tana MCP?

All three target outliner-style knowledge bases. Roam has the longest backend API tradition; Logseq is local-first OSS; Tana is AI-native and exposes its own integration layer.

Sources

Roam Research backend API — accessed 2026-04-20
Model Context Protocol — accessed 2026-04-20
MCP servers repo — accessed 2026-04-20