Documentation for agents
Read task-scoped Markdown, source manifests, and the read-only documentation MCP service.
Use a specific task page and its package contract before loading the whole corpus. Human pages and machine-readable pages are generated from the same content tree.
| Format | Endpoint | Use |
|---|---|---|
| Discovery | /llms.txt | Page titles, descriptions, and canonical URLs |
| Full corpus | /llms-full.txt | Whole-site context when required |
| Page Markdown | Append .md to a page path | Small, task-scoped context |
| Skill | /skill.md | Documentation workflow and boundaries |
| Page manifest | /reference/index.json | Machine-readable page discovery |
| Package coverage | /reference/coverage.json | Source manifests, versions, entry points, and page mapping |
| MCP | /api/mcp | Read-only page listing, retrieval and search |
MCP connection
Connect a Streamable HTTP MCP client to https://docs.openagent.id/api/mcp. The documentation service provides list_pages, get_page, and search. It only retrieves documentation. It does not generate identity keys, submit transactions, request credentials, or invoke product APIs.
Use the source boundary
- Match SDK language, manifest version, and enabled features.
- Treat an unverified registry release as source installation, not a public package promise.
- Keep standalone Rust and TypeScript HTTP adapter sessions separate.
- Configure identity, credential, and authority providers explicitly.
- Never infer cross-language encryption-key restoration from ACT vectors.
- Do not include secrets, seeds, sessions, private keys, or bearer credentials in documentation searches.
Begin with verification, authentication, or the library catalog.