The Model Context Protocol lets you connect external tools to Cogitator. GitHub, databases, file systems, search engines. Add a tool, and the agent discovers what it can do and figures out when to use it. No coding required.
Your agent is smart but limited. It can't access your GitHub repos, query your database, or interact with tools that aren't built in. Every new capability requires a software update.
Point Cogitator at a tool server. It could be a GitHub integration, a database connector, or anything that speaks the standard protocol. Configuration takes a few lines in the dashboard.
Once connected, Cogitator scans what tools the server provides. It reads their descriptions and learns when each one is appropriate to use. You don't need to map commands or write instructions.
When you ask something that needs an external tool, the agent calls it. "Create an issue in our repo" triggers the GitHub tool. "Look up that customer" queries your database. No special commands needed.
Servers only run when needed, so they don't consume resources when idle. The first request that needs a tool starts the server automatically.
If a server crashes, the next request restarts it automatically. You don't need to monitor processes or restart things manually.
Add, remove, start, stop, and test servers from the web interface. See which tools each server provides and whether it's currently running.
Uses the same configuration format as other AI tools, so existing setups transfer over. If you've configured a tool server elsewhere, you can reuse that configuration here.