Skip to content

code-review-graph.md

Latest commit

 

History

History
58 lines (47 loc) · 2.1 KB

code-review-graph.md

File metadata and controls

58 lines (47 loc) · 2.1 KB
Raw

Optional code-review-graph integration

Junto can use source context produced by an independently configured code-review-graph MCP server. The integration is optional: Junto does not install Python, download packages at runtime, manage the graph database, or start another MCP server.

Configure the graph server

Install and configure code-review-graph for Claude Code using its own documentation. To reduce MCP schema overhead, expose only the tools Junto commands can use:

{
  "mcpServers": {
    "code-review-graph": {
      "command": "code-review-graph",
      "args": [
        "serve",
        "--tools",
        "semantic_search_nodes_tool,get_review_context_tool,traverse_graph_tool,get_impact_radius_tool,detect_changes_tool"
      ]
    }
  }
}

Build the graph before starting a Junto task and keep it current using the graph tool's supported update or watch mechanism. Follow the host project's instructions when a repo_root is required.

Allow source context to leave the project

Junto consults may call remote APIs or a CLI that uses a remote model. Source-derived context is therefore disabled by default. Enable it only when every configured advisory backend is permitted to receive that source:

{
  "schemaVersion": 1,
  "gates": {},
  "consultContext": {
    "enabled": true,
    "maxChars": 12000,
    "persist": true
  }
}

maxChars is capped at 50,000. A lower value reduces repeated panel input. With persistence enabled, Junto writes the exact bounded snapshot to .junto/tasks/<id>/contexts/ and references it from each consult result. Junto marks this material as untrusted project data so source comments are not treated as model instructions. Context remains advisory and never satisfies a quality gate.

Fallback behavior

/junto:plan and /junto:panel try graph tools only when Claude Code exposes them. If the server is missing or a corrected query remains empty, the commands continue with native inspection or without attached panel context. The task lifecycle and quality gates do not depend on the graph integration.