fluree mcp

June 24, 2026 · View on GitHub

Model Context Protocol (MCP) server for IDE agent integration.

One fluree MCP server exposes a selectable set of toolsets over a single stdio transport, instead of one server per feature:

Toolset	Tools	Notes
`memory`	`memory_add`, `memory_recall`, `memory_update`, `memory_forget`, `memory_status`, `kg_query`	Developer-memory store; created lazily on first use
`docs`	`docs_search`, `docs_get`, `docs_examples`, `docs_tree`	Embedded, version-pinned documentation; stateless

Usage

fluree mcp <COMMAND>

Subcommands

Command	Description
`init`	Register the Fluree MCP server (with selected toolsets) in an IDE's config
`serve`	Start the MCP server (spawned by the IDE)
`status`	Show which toolsets are installed for each detected IDE

fluree mcp init

Write an IDE's MCP config so its agent spawns fluree mcp serve on demand. Needs no .fluree/ directory — the server lazy-inits its store and the docs toolset is stateless.

fluree mcp init [--ide <IDE>] [--toolsets <TOOLSETS>]

Options

Option	Description
`--ide <IDE>`	Target IDE: `claude-code`, `vscode`, `cursor`, `windsurf`, `zed` (auto-detected if omitted)
`--toolsets <TOOLSETS>`	Which toolset(s) to enable: `memory`, `docs`, a comma-separated list, or `all` (default)

# Register both toolsets with the auto-detected IDE
fluree mcp init

# Docs only (handy in a project that doesn't use Fluree Memory)
fluree mcp init --ide cursor --toolsets docs

init writes a single MCP entry named fluree whose --toolsets arg records the selection, and removes any legacy per-feature fluree-memory / fluree-docs entries from a previous install. After it runs, reload your editor to activate the tools.

The store is created lazily on first tool call, so there is no separate initialization step. To git-track repo memories, commit .fluree-memory/ after first use.

fluree mcp serve

Start the MCP server. Normally the IDE spawns this; run it by hand only to debug.

fluree mcp serve [--transport <TRANSPORT>] [--toolsets <TOOLSETS>]

Options

Option	Description
`--transport <TRANSPORT>`	Transport protocol: `stdio` (default)
`--toolsets <TOOLSETS>`	Which toolset(s) to expose. Defaults to `memory` for back-compat when omitted; `init` always writes an explicit `--toolsets`.

The stdio transport reads JSON-RPC requests from stdin and writes responses to stdout. CLI tracing is disabled (or written to .fluree-memory/.local/mcp.log when the memory toolset is active) so nothing pollutes the JSON-RPC stream on stdout/stderr.

A docs-only server (--toolsets docs) is fully stateless and requires no .fluree/ directory.

Manual IDE configuration

fluree mcp init writes this for you, but to add it by hand:

{
  "mcpServers": {
    "fluree": {
      "type": "stdio",
      "command": "/path/to/fluree",
      "args": ["mcp", "serve", "--transport", "stdio", "--toolsets", "memory,docs"],
      "env": {
        "FLUREE_HOME": "${workspaceFolder}/.fluree"
      }
    }
  }
}

The env.FLUREE_HOME is only needed when the memory toolset is enabled (it pins the store to the workspace). For the per-IDE config file paths and keys, see Memory: IDE support matrix.

Testing with JSON-RPC

printf '%s\n' \
  '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"smoke","version":"0.0"}}}' \
  '{"jsonrpc":"2.0","method":"notifications/initialized","params":{}}' \
  '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' \
  | fluree mcp serve --transport stdio --toolsets all

fluree mcp status

Report, per detected IDE, whether the fluree server is installed and which toolsets it exposes.

fluree mcp status

Back-compat

These older commands still work as hidden aliases:

Old	Now equivalent to
`fluree mcp install [--ide …] [--toolsets …]`	`fluree mcp init …`
`fluree memory mcp-install [--ide …]`	`fluree mcp init --toolsets memory`
`fluree memory init`	`fluree mcp init --toolsets memory` (store is no longer created up front)