Cursor integration

What it is¶

The CodeClone MCP integration for Cursor enables AI-powered code analysis and change control directly in the editor. Through Cursor's MCP (Model Context Protocol) server support, you can run CodeClone analysis, review patch contracts, and track engineering memory without leaving your IDE.

When to use it¶

Use the Cursor integration when you:

Need live structural feedback on Python code before committing
Want change control gates on risky patches (high complexity, coupling, or clones)
Track and recover from previous analysis decisions via engineering memory
Review PR quality across your team using standardized baselines
Debug reproducible findings (clones, dead code, structural issues)

Basic workflow¶

graph LR
    A["Open Python file<br/>in Cursor"] --> B["Trigger analyze<br/>via MCP"]
    B --> C["View findings<br/>in sidebar"]
    C --> D["Select finding<br/>for detail"]
    D --> E["Review remediation<br/>hints & memory"]
    E --> F{Ready to<br/>commit?}
    F -->|No| G["Edit & re-analyze"]
    G --> B
    F -->|Yes| H["Run change control"]
    H --> I["Accept patch<br/>or revise"]

Key commands¶

The integration exposes these MCP tools through Cursor's interface:

Command	Purpose	When to use
`analyze_repository`	Full structural analysis from repo root	First pass, before major refactoring
`analyze_changed_paths`	Focused review of changed files only	During PR review, local branches
`start_controlled_change`	Declare intent, compute blast radius	Before editing large chunks
`finish_controlled_change`	Verify patch, record trail, accept/reject	After editing, before commit
`get_implementation_context`	Bounded structural & call-graph facts	Planning edits, understanding scope
`get_production_triage`	Health snapshot with hotspots	Quick health check

Install¶

First install the CodeClone MCP runtime:

uv tool install --prerelease allow "codeclone[mcp]"

The Cursor plugin bundles the launcher and configures MCP for you. To configure it manually instead, add a .cursor/mcp.json in your project that runs the installed codeclone-mcp entry point over stdio:

{
  "mcpServers": {
    "codeclone": {
      "command": "codeclone-mcp",
      "args": ["--transport", "stdio"]
    }
  }
}

The plugin's own launcher invokes python3 ./scripts/launch_mcp.py; the manual config above is the equivalent when you have codeclone[mcp] installed on your PATH.

Bundled skills, rules, and hooks¶

The Cursor plugin ships:

10 skills — the same set as the Claude integration (change control, review, triage, blast radius, implementation context, hotspots, engineering memory, setup, architecture triage, platform observability).
3 rules (.mdc) — change-control-gate, codeclone-python, codeclone-workflow.
Change-control hooks — a pre-tool-use gate and a post-edit hook that enforce the intent-first workflow around Python edits.

Common mistakes¶

Mistake 1: Absolute paths in MCP calls MCP tools require absolute repository roots. Relative paths like . are rejected. Always pass the full workspace path.

Mistake 2: Forgetting start_controlled_change before editing Declaring intent first ensures the change control workflow captures blast radius and budget. Skipping it makes finish_controlled_change unable to verify scope.

Mistake 3: Mixing before and after analysis runs For Python structural changes, you must run analyze_repository before editing and again after. A single run cannot verify both states.

Mistake 4: Ignoring stale memory warnings Engineering memory may reference outdated decisions or contradictions. Always review memory alerts before relying on recorded facts.

Next steps¶

Read Engineering Memory to understand how CodeClone tracks decisions across sessions
Explore MCP tool details with the in-tool help(topic="overview") call, or the MCP tools reference
Set up baseline thresholds in pyproject.toml to match your team's standards
Use generate_pr_summary(format="markdown") to produce a compact PR summary — see the MCP tools reference