Read workflow truth
Inspect runs, Workflow Cases, approvals, compare outputs, proof packs, and incident summaries without mutating anything.
Read-only MCP quickstart.
CortexPilot already exposes a real Model Context Protocol (MCP) surface, but the public contract is intentionally read-only. This page exists so teams searching for MCP, Codex, Claude Code, or OpenClaw adoption paths can find the truthful protocol story quickly instead of guessing from source files.
Use this page when your first question is: “What can external tools safely inspect right now?” If your first question is instead about package imports, builder bootstrap, or repo-owned skills, this page should route you to the better next click.
What the current MCP surface is for
Return structured JSON
Expose a repo-owned protocol boundary that other tools and agents can consume safely instead of screen-scraping dashboard pages.
Stay below execution authority
Keep operator truth readable without inventing a second control plane, a hosted account layer, or write-capable MCP claims.
Fast adoption path
| My first question | Start here | Why |
|---|---|---|
| I need the shortest truthful protocol boundary | This page | Use MCP first when your team needs machine-readable inspection before packages or builder code. |
| I need the “which adoption ladder fits us?” answer | Compatibility matrix | Use it when the main question is whether you should start from MCP, skills, builders, or proof-first cases. |
| I also need HTTP, types, or a thin JS client | API and contract quickstart | Use the API page when you want the route/type/client edge in addition to the protocol story. |
| I need repo-owned builder packages | Builder quickstart | Use it when you are ready to import package-level helpers rather than only understand the protocol boundary. |
| I need coding-agent team guidance | Integration guide and skills quickstart | Use these when your real question is how a Codex / Claude Code / OpenClaw team should adopt the repo truthfully. |
Try the current server
The current server is a repo-owned stdio JSON-RPC node. Run it from the repo root when you want the shortest proof that the public MCP story is real:
PYTHONPATH=apps/orchestrator/src python3 -m cortexpilot_orch.cli mcp-readonly-server
Current tool set:
list_runs,
get_run,
get_run_events,
get_run_reports,
list_workflows,
get_workflow,
list_queue,
get_pending_approvals,
get_diff_gate_state,
get_compare_summary,
get_proof_summary,
and get_incident_summary.
What it does not do yet
- no write-capable MCP mutations
- no hosted operator account or managed login layer
- no claim that MCP replaces the orchestration runtime itself
Why this matters for different audiences
Protocol-first inspection
Useful when a Codex-facing workflow needs one machine-readable place to inspect workflow truth before promoting package or skills adoption.
MCP-compatible workflow context
Useful when a Claude Code team wants the protocol layer to stay honest about what can be read today without implying write-enabled MCP.
Adjacent protocol layer
Useful as a read-oriented integration boundary around OpenClaw workflows, while plugin-ready wording stays separate from the current MCP story.
Where to read the current contract
- read-only MCP server implementation
- orchestrator README
- MCP architecture baseline
- AI + MCP + API surfaces for the broader product map