MCP Server

Note

The EOS MCP server is completely optional and does not provide any functionality beyond the interactive Python interface.

eos-mcp-server exposes EOS’s observables, parameters, constraints, and references as read-only lookup tools for an MCP client (Claude Code, Claude Desktop, Codex, …), so an LLM coding agent can look up correct qualified names, units, and kinematics instead of guessing. It runs locally over stdio; it performs no computation and never modifies EOS state.

Installing

After building EOS with ./configure --enable-python --prefix=<venv> and make install (see Build and Install from Source), install the mcp SDK into the same environment:

<venv>/bin/pip install mcp

<venv>/bin/eos-mcp-server is then ready to use.

Connecting a client

Point the client at <venv>/bin/eos-mcp-server, with EOS_SOURCE_DIR set to your checkout root (needed only to serve the two doc resources below; the lookup tools work without it).

Claude Code (project .mcp.json, or claude mcp add):

{ "mcpServers": { "eos": {
  "command": "<venv>/bin/eos-mcp-server",
  "env": { "EOS_SOURCE_DIR": "/path/to/eos" }
} } }

Codex CLI (~/.codex/config.toml, or codex mcp add):

[mcp_servers.eos]
command = "<venv>/bin/eos-mcp-server"
env = { EOS_SOURCE_DIR = "/path/to/eos" }

Tools and resources

observables, parameters, constraints – list entries matching a prefix/name/suffix filter, or describe a single one given its qualified_name.

references – list entries matching a year/index filter, or describe a single one given its key.

eos://doc/cli-reference, eos://doc/defining-observables – the corresponding reference documents.