MCP trigger
AI agents and MCP tools, natively
hostfunc is the first function platform built for agents. Its MCP server exposes your functions as tools any LLM can call — list, read, write, run, debug — using the same handlers the dashboard uses. Agents can even create scratch functions that auto-delete on a TTL.
The problem
Giving an LLM real capabilities means writing and hosting tools, then wiring an MCP server, auth, rate limits, and an audit trail. Building that glue layer is most of the work — and it's the same every time.
How hostfunc solves it
- Built-in MCP server at /api/mcp — paste a token into Claude Desktop and the model has 9 tools instantly.
- Scratch functions: the agent creates a one-off function that returns a result and self-destructs on a TTL.
- Every fn-to-fn call is recorded, so you can watch an agent compose your graph in real time.
- Same auth, rate limits, and audit trail across the dashboard, CLI, and MCP — no separate surface to secure.
claude_desktop_config.json
{
"mcpServers": {
"hostfunc": {
"url": "https://hostfunc.io/api/mcp",
"headers": { "Authorization": "Bearer hf_live_…" }
}
}
}