Set up the callgraph MCP server

Before you start

The MCP server is the same binary the skill uses, so install the callgraph package first. The binary is in the skill package: ~/.claude/skills/callgraph/bin/<os>-<arch>/callgraph (folders: darwin-arm64, darwin-x86_64, linux-x86_64, linux-aarch64; Windows: bin\windows-x86_64\callgraph.exe).

JSON configs need the absolute path (~ is not expanded there). This prints it on macOS / Linux:

echo ~/.claude/skills/callgraph/bin/$(uname -s | tr '[:upper:]' '[:lower:]')-$(uname -m)/callgraph

Not every client starts the server in your project folder, so register your projects as aliases once:

bash ~/.claude/skills/callgraph/projects.sh add shop /absolute/path/to/shop

Configure your client

Every example uses --auto-update (recommended, see updates). Replace /Users/you/…/darwin-arm64/callgraph with the path printed above.

claude mcp add -s user callgraph -- \
  ~/.claude/skills/callgraph/bin/$(uname -s | tr '[:upper:]' '[:lower:]')-$(uname -m)/callgraph mcp --auto-update

claude mcp add -s user callgraph -- "$env:USERPROFILE\.claude\skills\callgraph\bin\windows-x86_64\callgraph.exe" mcp --auto-update

{
  "mcpServers": {
    "callgraph": {
      "command": "/Users/you/.claude/skills/callgraph/bin/darwin-arm64/callgraph",
      "args": ["mcp", "--auto-update"]
    }
  }
}

{
  "mcpServers": {
    "callgraph": {
      "command": "/Users/you/.claude/skills/callgraph/bin/darwin-arm64/callgraph",
      "args": ["mcp", "--auto-update"]
    }
  }
}

{
  "servers": {
    "callgraph": {
      "type": "stdio",
      "command": "/Users/you/.claude/skills/callgraph/bin/darwin-arm64/callgraph",
      "args": ["mcp", "--auto-update"],
      "cwd": "${workspaceFolder}"
    }
  }
}

Windows paths in JSON

On Windows the binary is callgraph.exe under bin\windows-x86_64\. In JSON every \ is doubled; replace you with your Windows user name:

{
  "mcpServers": {
    "callgraph": {
      "command": "C:\\Users\\you\\.claude\\skills\\callgraph\\bin\\windows-x86_64\\callgraph.exe",
      "args": ["mcp", "--auto-update"]
    }
  }
}

For VS Code, use the same command value in the servers entry above.

The ten tools

Paths are relative to the project root: a registered alias (project) or the server's working directory. A typical flow: locate → outline → method_body → find_callers.

Optional layer parameters (the default output does not change): layer on outline, search and find_callers keeps only the files of that layer; by_layer on find_callers orders callers by layer. See project profile.

On connect the server sends the client a short usage guide in your language (which tool to use when, how to find callers, how to read the output); Claude Code puts it into the agent's instructions. If the server starts in a project folder, a compact project profile is appended.

What runs locally

The only network call is the update check below: at most once a day it looks up the latest GitHub release (and, with --auto-update, downloads it).

Updates: --auto-update

• With --auto-update (recommended): at start, at most once a day, the server checks the latest release. If there is a newer one, it downloads it, verifies it against the release's SHA256SUMS (nothing is installed on a mismatch or a missing checksum), installs it exactly like callgraph.sh update (your projects and settings are kept) and restarts itself, so the client talks to the new version right away.
• If anything fails (no network, rate limit, bad checksum, no write permission), it writes one line to stderr — callgraph: auto-update skipped, continuing with …, which clients log — and starts the current version.
• Without the flag: the server checks in the background and adds one line to the first answer of the session: ℹ callgraph X is available (installed: Y) — update: ….
• It writes only into callgraph's own folder, never into your projects.
• Turn all update checks off with CALLGRAPH_UPDATE_CHECK=0 in the server's environment (the env field of the client config): then there is no network request at all. Checks are also off in CI.

Troubleshooting

Related