joystream-ai/joystream
Build, ship and debug JoyStream agents through the JoyStream MCP server.
Write or change a JoyStream agent spec through the JoyStream MCP server — turn what the user wants into a valid definition, choose skills and connected services, find the connector actions it needs, decide whether it runs code in the sandbox (files, conversions, audits, data crunching) and wire that up, shape its output so every run saves it as a file, and fix validator errors — then hand off to joystream-agent-ship.
Diagnose one of your own failed JoyStream agent runs through the JoyStream MCP server — read the run, its failed ticket and the agent version that ran, find the root cause, and propose a spec change, or say plainly that the cause is unclear.
Take a JoyStream agent from a spec change to a real run through the JoyStream MCP server — update, check readiness, build a version, run a Simulation of it, then a real Run — stopping for the user's approval before every real Run. Also use it to run an agent and report what the run did, whenever a run is started from this client.