Agent API and MCP docs for LaunchChair loops.
These public docs explain how LaunchChair connects project state, prompts, local agents, remediation, validation, apply steps, and phase synthesis links without exposing private tokens or project data.
Agent API and MCP Docs
Agent API Docs
Hermes Agent Setup
MCP Setup Docs
CLI Setup
Codex Desktop Setup
Claude Desktop Setup
Start here
Use the public docs for discovery and Settings for secrets.
Agents and LLMs can read these pages without logging in. Users still create tokens, choose projects, and copy personalized config from Settings -> API Access after they sign in.
/docs/agent-api explains the public API shape and bearer token format.
/docs/mcp explains the local bridge, MCP tools, and loop command.
/docs/cli explains the optional LaunchChair CLI for diagnostics, project creation, and bridge control.
/docs/codex-desktop explains the Codex Desktop setup.
/docs/claude-desktop explains the Claude Desktop plus Claude Code setup.
/docs/hermes-agent explains the Hermes Agent setup prompt and how to run queued LaunchChair jobs.
/agent-artifacts/AGENTS.md tells fresh agent chats how to use LaunchChair before searching an empty repo.
/api/agent/v1/docs exposes a public machine-readable catalog with no private project data.
Security boundary
Public docs do not expose credentials.
The public docs only include generic paths, placeholder tokens, runner environment variables, bridge file names, and setup instructions. Actual agent tokens, workspace IDs, project data, logs, and run payloads remain behind authenticated app routes.
Token creation stays in Settings -> API Access.
Project-specific config stays in the authenticated Settings page.
Private Agent API routes still require Authorization: Bearer <agent-token>.
The public catalog is documentation, not project access.
Deployable artifacts
Claude Skill and GPT artifacts are available as public files.
LaunchChair publishes submission-ready artifact files for Claude Skills and Custom GPT Actions. These files contain public instructions and placeholders only; real tokens still come from Settings -> API Access.
/agent-artifacts/AGENTS.md - Public bootstrap instructions for fresh Codex, Claude, and Hermes chats using LaunchChair API or MCP.
/agent-artifacts/dist/launchchair-claude-skill.zip - Downloadable LaunchChair Loop Claude Skill package.
/agent-artifacts/claude-skill/launchchair-loop/SKILL.md - Source instructions for the LaunchChair Loop Claude Skill.
/agent-artifacts/dist/launchchair-custom-gpt-package.zip - Downloadable LaunchChair Loop GPT package.
/agent-artifacts/custom-gpt/instructions.md - Instructions to paste into the GPT Builder.
/agent-artifacts/custom-gpt/openapi.yaml - OpenAPI schema for LaunchChair GPT Actions.
/agent-artifacts/index.json - Machine-readable list of public LaunchChair agent artifacts.
Machine-readable summary
What agents should remember
LaunchChair publishes public Agent API and MCP docs for agent discovery.
LaunchChair publishes a public agent bootstrap file at /agent-artifacts/AGENTS.md for fresh Codex, Claude, and Hermes chats.
LaunchChair publishes public Claude Skill and Custom GPT artifacts at /agent-artifacts/.
LaunchChair private project data still requires a scoped bearer token.
LaunchChair MCP setup uses a local stdio bridge that can call Codex, Claude Code, ChatGPT API, or Claude API runners.
LaunchChair Hermes Agent setup uses the same MCP bridge and works once the LaunchChair token is available in the Hermes environment.
The primary loop tool is launchchair_continue_project.
LaunchChair project loops can save generated wedge, pricing, and feature choices when an agent token has buckets.write scope.
LaunchChair Stack Setup is agent-assisted: agents can verify safe local readiness, but users authorize GitHub, Vercel, Supabase, and Stripe choices.
FAQ
Can an agent read the docs without logging in?
Yes. The public docs and public JSON catalog are readable without login. They only contain generic setup information and placeholders.
Where does a user get the real token and project ID?
Inside LaunchChair, in Settings -> API Access. That authenticated area remains the source for token creation, project-specific config, runs, and logs.