Test AI agent webhooks & tool-callbacks

When an agent calls a tool or fires a callback, you need to see the real request it sent. Hand it a Catchr URL and read back precisely what it did — headers, auth token, and payload included.

• Agent-native. A bin is created and read with plain HTTP, so an agent can provision and verify its own callbacks — see /llms.txt and /openapi.json.
• Full fidelity. Inspect the exact method, headers (including auth), query and body the agent sent.
• Zero setup. No account, no keys to manage beyond the per-bin token you get on create.
• Recipes ready. Copy-paste pytest / Node / CI snippets to assert on captured calls.

# one call, no signup
curl -s -X POST https://catchr.perch-app.workers.dev/api/bins
# → {"id":"a1b2c3d4","url":"https://catchr.perch-app.workers.dev/h/a1b2c3d4","token":"…"}
# send anything to the url, then read it back:
curl "https://catchr.perch-app.workers.dev/api/bins/a1b2c3d4/requests?token=…"

See copy-paste CI & agent recipes →

Why agents need a request inspector

When an AI agent calls an external tool or fires a callback, the only way to know it did the right thing is to see the actual HTTP request it produced — the endpoint, the auth header, the JSON body. A Catchr bin is a safe target for exactly that: point the agent's tool at the bin URL, let it run, then read back the precise request it sent.

Two ways to wire it up

1. As a plain HTTP target. Give the agent a bin URL, then GET /api/bins/:id/requests?token=…&wait=25 to block until the call lands and assert on it — no polling loop.
2. As an MCP server. Add Catchr's remote MCP endpoint (/mcp) to Claude, Cursor or any MCP client, and the agent gets create_bin, get_requests and set_response tools so it can provision and verify its own callbacks.