CCAPI

Chinchilla's shared internal Claude Code API — one always-on endpoint, any app gets Claude's agentic tools over plain HTTP.

Why this exists

No per-app Claude wiring, no spawning your own claude processes. Any product or agent gets web-search-backed answers over a single HTTP call.

Endpoints

Method	Path	What it does
GET	`/`	This page.
GET	`/health`	Liveness check — `{"ok": true}`
POST	`/query`	Run a prompt through Claude Code (web search + fetch enabled). Returns the final answer text.

Usage

curl -s https://ccapi.chinchilla-ai.com/query \
  -H 'content-type: application/json' \
  -d '{"prompt":"Search the web for X and answer concisely"}'

Response:

{"answer": "<final result text>", "duration_ms": 8421}

Latency is typically 5–20s — it does real agentic web search/fetch work per request, not a cached lookup.

What it can do

Runs Claude Code's query() with maxTurns: 5, tools WebSearch and WebFetch only, no MCP servers, no file/shell access. It's a read-only research tool, not a general agent.

Auth & access

This is an internal endpoint — no public auth yet (Phase 1). Keep the URL out of anything public-facing until a token ships.

Status

Runs on a dedicated Graviton EC2 box (ccapi.chinchilla-ai.com) as of 2026-07-09 — migrated off the shared MacBook Air Lima VM it used to share space on. See team-workspace/infra/ccapi/ for full infra docs.