MCP for AI Agents¶
Use CodeClone through codeclone-mcp — same pipeline and report as the CLI.
Analysis truth is read-only: MCP never mutates source, baselines, analysis cache, or canonical reports. It may write session-local coordination (workspace intents), Engineering Memory drafts, and optional audit rows when enabled. Opt-in Platform Observability writes separate local development telemetry and never becomes repository truth.
Install: Getting started — MCP extra.
Guide vs contract
This section is how to work with MCP. Tool names, parameters, and response shapes are normative in the MCP interface contract.
Setup¶
| Step | Page |
|---|---|
| Register a client | Client setup |
| Launcher & transport | Server & transport |
| Layer diagram | Architecture |
| Troubleshooting & issues | MCP troubleshooting |
Workflows (recommended order)¶
| Phase | Recipe |
|---|---|
| 1. Baseline-aware triage | Analyze & triage |
| 2. Focused inspection | Drill down & checks |
| 3. Live code context | Analyze & triage |
| 4. Governed edits | Change control |
| 5. Durable scope context | Memory recipes |
| 6. Optional: coverage & session markers | Coverage join & session markers |
Maintainers only (developing CodeClone itself — not user repo review):
| Phase | Recipe |
|---|---|
| M. Observer setup & MCP drill | Platform Observability recipes |
| M. Maintainer playbook | Developing CodeClone with observer |
Reference shortcuts¶
| Need | Page |
|---|---|
| Prompt patterns | Prompt patterns |
| Payload field cheat sheet | Payload cheatsheet |
| Change control contract | Structural Change Controller |
| Implementation-context contract | Implementation context |
help() topics |
Help topics |
| Engineering Memory contract | Engineering Memory |
| Runtime diagnostics (maintainer-only) | Platform Observability |