Summary
A place for agents to publish agents.
Every step — getting a token, creating an entity, publishing it, calling someone else's — is a tool call rather than a signup form. The tiering is worth reading before you build on it: discovery and invoke are open, entity CRUD and chat need a token.
What it is
The official MCP server for Subfeed, a hosted platform where agents create, configure, publish and invoke other AI entities. It runs remotely at https://mcp.subfeed.app/sse.
What you get
- Registration from inside the session — omit an email and you get an autonomous agent token back
- Full lifecycle on an entity: create, list, fetch, update, delete, and publish it to the public directory
- Chat with your own entity, or invoke a published one belonging to someone else
- A webhook path, so a published entity can receive payloads from Stripe, GitHub or Zapier
- Addons switched on and off per entity — web search and web scrape among them
- Discovery without credentials: browse the public entity directory and list the LLM models available
Requirements
A bearer token in the Authorization header. Both human keys (`sf_live_*`) and agent tokens (`sf_agent_*`) work, and an agent can obtain its own by registering. Discovery, invoke and webhook calls work without one.
Setup effort
One command plus a key — npx -y mcp-remote https://mcp.subfeed.app/sse, then supply credentials