Chapter 22 — Paperclip adapter (retired)

This chapter was removed in v0.110.0. It documented @thclaws/paperclip-adapter, an npm package that let a Paperclip orchestration hire a thClaws agent as one of its runtimes. The product line it served has been retired, and the adapter’s source no longer ships in this repository.

The page is kept as a stub so existing links don’t break.

If you came here to drive thClaws from another system

That still works — it was never specific to Paperclip. thClaws exposes two HTTP surfaces under thclaws --serve, and any orchestrator, scheduler, or CI job can use them:

• POST /agent/run — the thClaws-native shape. Takes a prompt and an optional workspace_dir, runs the full skill / MCP / plugin / policy bootstrap scoped to that directory, and streams native events (tool calls, skill invocations) rather than pretending to be OpenAI tokens. Supports multi-turn continuation via session_id and fire-and-forget delivery via x_callback.
• POST /v1/chat/completions — OpenAI-compatible, for clients that already speak that protocol (Cursor, Aider, n8n, openai-python).

GET /v1/agent/info reports what a running daemon has available — skills, MCP servers, model catalogue, version — so an orchestrator can show capabilities it never pushed itself.

Full reference: agent-endpoint.md, openai-api.md, and agent-info-endpoint.md in the technical manual.

If you were using the npm package

It remains published on npm and installable, but it is unmaintained and no longer built or tested from this repository. Point your integration at POST /agent/run instead — it is the same capability without the wrapper.