Summary
HybridClaw has strong MCP client support and its own internal web/API surfaces, but it does not appear to expose a first-class server-side integration surface for external editors or agent clients such as an MCP server endpoint or a standardized agent API.
Why
This would let other tools treat HybridClaw as a reusable agent runtime instead of only a standalone CLI/gateway product. It would also make editor and automation integrations much easier.
Proposed scope
- Evaluate two target surfaces:
- HybridClaw as an MCP server
- HybridClaw as a standardized external agent API (for example OpenAI-style chat/responses-compatible subset)
- Choose one as the first shipped interface and define the compatibility boundary clearly.
- Expose core operations such as send message, stream response, list/search sessions, inspect agents, and optionally manage attachments.
- Define auth and permission boundaries for remote clients.
Candidate MVP
- HTTP MCP server endpoint or stdio MCP mode backed by the existing gateway runtime
- tools/resources for conversations, sessions, agents, and message submission
- token-protected API surface with explicit scope controls
Implementation notes
- Do not mirror every internal concept on day one; start with a narrow, stable contract.
- Reuse existing gateway auth, session routing, and audit primitives where possible.
- Treat remote clients as untrusted and preserve the current approval/security model.
- Document the compatibility story clearly so users know whether the goal is MCP-first, OpenAI-compatible, or both.
Acceptance criteria
- External clients can connect to HybridClaw through one documented, supported server-side interface.
- The interface supports at least message submission and response retrieval/streaming.
- Auth, audit, and approval behavior are preserved for remote callers.
- The docs include setup examples and a supported capability matrix.
- Tests cover auth failures, basic request flow, and one end-to-end client integration path.
Summary
HybridClaw has strong MCP client support and its own internal web/API surfaces, but it does not appear to expose a first-class server-side integration surface for external editors or agent clients such as an MCP server endpoint or a standardized agent API.
Why
This would let other tools treat HybridClaw as a reusable agent runtime instead of only a standalone CLI/gateway product. It would also make editor and automation integrations much easier.
Proposed scope
Candidate MVP
Implementation notes
Acceptance criteria