Subcore AI documentation
Developer documentation for the Subcore AI platform, covering the telemetry read API, the agents API and the MCP server.
Subcore runs conversational AI agents for customers, teams and operations, and records everything they do. These pages document the three surfaces you talk to from your own code.
Telemetry read API
Read back the conversations your agents have had: sessions, events and evaluation runs, over HTTPS.
Agents API
Register your agents and store the prompt files they answer with, then see what a running deployment reads back on every call.
MCP server
Connect Claude or any MCP client to your agents: run test turns through them and query the same telemetry.
What you need
All three surfaces live on one host and take the same kind of credential.
| Base URL | https://api.subcore.ai |
| Credential | A Subcore API key, sent as Authorization: Bearer <key> |
The key identifies one organization, and every result is scoped to it. There is no account id, tenant header or organization parameter anywhere in any of the three: the key is the tenant. Create and revoke keys in the Subcore console. Managing agents also needs the manage-agents permission on the key, which an admin in your organization grants in the console.
The one concept to read first
A session is every event that shares a session id (sid). There is no
session object that gets created and closed. A conversation exists because
several events carry the same id, and reusing an id later appends to that same
conversation rather than starting a new one.
This shapes both telemetry surfaces, so it is worth reading once in full:
- In the read API, sessions are a view over events, not a table you can create rows in.
- In the MCP server, session semantics are why you omit the session id to start a conversation and pass it only to continue one.
What is not here
This site documents api.subcore.ai. The Subcore kit that runs inside your own
deployment, and the Subcore console, are documented separately.