app.cotera.co

Cotera is where teams build AI agents on top of their own data and tools. This connector puts that workspace inside Claude: browse the agents you already have, build new ones, run them, and bring the output back into the conversation.

Build agents. Describe what an agent should do in plain markdown and Claude creates it in your workspace, with headings, lists, and code blocks intact. Prompts can reference your connected tools, your documents, and your other agents, so a new agent delegates work the same way one built in the Cotera editor does.

Find what to reference. Search the tools your workspace has connected — including whether each one is currently connected and ready to run — and search your documents, so an agent is wired to the real thing rather than to a guess.

Run them, one input or many. Run an agent against a single input, or start up to 50 runs in parallel over a list: a batch of leads, tickets, transcripts, or rows. Runs execute in the background, and you collect the outputs as they finish.

Read what you have. List every agent in the workspace, newest first, including ones your teammates built in the Cotera app, and read any agent's full prompt.

Everything happens in the workspace you authenticate with. Agents created here are ordinary Cotera agents — your team can open, edit, schedule, and share them in the app afterwards. Nothing is destructive: creating an agent never modifies an existing one, and running an agent never changes it.

Two things I deliberately kept in, since they're the details that make the listing trustworthy: the 50-run batch ceiling (the real MAX_BATCH_SIZE) and the read-only/non-destructive framing, which matches the tool annotations you ship.