Intellij - Github copilot

[hari-prasanth-maersk]

Hi, I tried to configure a mcp server for GitHub copilot. MCP server is running but sometimes i feel code-review-graph is not reliable while seeing the graph nodes.

Any idea? If i missed any important config?

[tirth8205]

There is no IntelliJ-specific install target, which is probably the root of this. code-review-graph install --platform copilot writes .vscode/mcp.json with a servers key (code_review_graph/skills.py, the "copilot" entry around line 249). That file is only read by Copilot in VS Code. The Copilot plugin in IntelliJ ignores it and reads its own global config instead, which you edit via Settings > GitHub Copilot > Model Context Protocol (on disk it lands under ~/.config/github-copilot/intellij/mcp.json).

The unreliability you describe usually comes down to working directory. Our VS Code entry sets cwd to the repo root so serve finds .code-review-graph/graph.db. The JetBrains config is global and does not carry a per-project cwd, so the server auto-detects the repo from wherever the IDE spawned it. Depending on that, it can resolve no graph or the wrong one. Pin it explicitly, since serve accepts --repo (cli.py line 1181):

{
  "servers": {
    "code-review-graph": {
      "command": "uvx",
      "args": ["code-review-graph", "serve", "--repo", "/abs/path/to/your/repo"]
    }
  }
}

Concrete checks:

1. Run code-review-graph status --repo /abs/path/to/your/repo in a terminal. If it reports no graph or zero nodes, the tools have nothing to return; run build first.
2. Pass an absolute repo_root in tool calls when in doubt. Every tool accepts it and it overrides auto-detection.
3. The first call after IDE start is slow because the server cold-starts. Later calls in the same session should be fast; if every call is slow, the plugin is likely respawning the process.

If you get it working, a note on which config path your IntelliJ version used would help others here.