AI agent integration matrix
Use this page to choose an integration path. It is written for assistant builders, MCP clients, and humans steering an agent who want the same vocabulary. Machines should still read /agent-tools.json or /mcp-tools.json.
Goals × integration mechanism
| Goal | Mechanism | Primary entry |
|---|---|---|
| Let the user run a transform in a browser tab | Prefilled HTTPS URL from manifest | /ai-agents, /agent-tools.json |
| Headless automation needs a JSON result (no browser) | Allowlisted POST execute | GET /api/agent/v1/capabilities, Server API matrix |
| Exclude AI-assisted catalog entries | Non-AI-assisted manifest | /mcp-tools.json |
| Stdio MCP in a compatible client | Local stdio MCP | bun run mcp — see MCP & Actions guide |
| Quick terminal-style deep links | Pseudo-CLI | /tools/cli |
| Generative explain / NL → regex, etc. | AI-assisted tool page + POST /api/ai/* | /tools/ai, Server route matrix |
Copy-paste instructions for assistants
GET /llm-prompt.txt mirrors the English instructions in the manifest. Pair it with Offload compute from LLMs when explaining why to hand work to Toolcore.
Common use cases
- Onboard a new teammate wiring assistant actions or MCP: send this matrix plus /ai-agents/server-api.
- Decide in one glance whether the next step is a prefilled URL or a POST to /api/agent/v1/execute.
Common mistakes to avoid
Skipping GET /api/agent/v1/capabilities before POST execute
The allowlisted operation IDs and body shape are versioned there; hard-coding URLs without refreshing capabilities can break when capabilities change.
FAQ
Is this page a substitute for the JSON manifests?
No. Crawlers and agents should still fetch /agent-tools.json or /mcp-tools.json for the full tool list and prefill rules. This page is a human- and retrieval-friendly map.