MCP quickstart
Connect an agent to Tensorant with Claude Code and a sign-in in your browser, then ask it about your projects.
An agent is an AI assistant that works inside an app such as Claude Code, Cursor, Codex, VS Code, Claude or ChatGPT. Once it is connected to Tensorant, you can ask it in plain words to do what you would do in the console: import a dataset, review examples, run an evaluation, train a model, test it and publish it. Agent apps call this kind of connection an MCP server.
This page gets you to a first answer with Claude Code. Connect your app has the setup for every app, and Tools is the full list of what an agent can do.
An agent acts as you, in one organization, with the access level you give it. It follows your role, your plan's limits and the same separation between organizations as the console, and what it changes appears in the audit trail. Every plan, Free included, can connect agents.
Agents are different from the API. An agent manages your projects, data, runs and models. Your applications call your published models through the API, with an API key.
Before you start
- An account in an organization. Viewers can connect agents too, at Read access.
- Claude Code, or another app that can use a remote server: see Connect your app for the others.
- To train or deploy later, your organization needs a GPU provider and storage connected by an owner in Connections. You do not need them for this page.
Step 1: Copy the address
The address agents connect to is:
https://mcp.tensorant.io/mcpSettings → Agents shows the same address, with Copy address.
Step 2: Add it to Claude Code
Run this in a terminal. Then type /mcp inside Claude Code and follow the steps to sign in in your browser.
claude mcp add --transport http tensorant https://mcp.tensorant.io/mcpStep 3: Approve it in your browser
Claude Code sends you to a Tensorant tab in your browser.
- If you are signed out, sign in. The request is still there afterwards.
- If you belong to several organizations, choose the Organization the agent works in. The page starts on the one you last worked in in the console on this browser; check that it is the right one.
- Under Access, choose Read, Build or Full. Apps often ask for all three levels, so the page usually starts on Read: choose more only if the agent needs it. See Access levels.
- Choose Allow. The page says "You allowed app". Go back to the app: Return to the app takes you there, and you can close the tab.
The agent now appears under Settings → Agents, with Browser sign-in under How. The full approval page, with every message on it, is described in Connect your app.
Step 4: Ask the agent something
Ask in plain words, such as "Which projects do I have in Tensorant, and what does my plan allow?" Agents are told to call get_account first. It shows who the agent acts for: you, the organization, your plan with its limits and this month's usage, and which connections are set up (never their values). list_projects lists your projects.
At Read access an agent only looks and changes nothing. You cannot change an agent's level after you connect it: to give it more, connect again at Build or Full, then revoke the old one. At Build it can do things such as "import the first 500 rows of this dataset into my support project". Training and deployments need Full, and the agent shows you the price first: see Spending.
When something goes wrong
| What you see | What to do |
|---|---|
| This app cannot connect: "This app is not known to Tensorant. Connect it again from the app." | Start the connection again from the agent app and approve the new request. |
| No organization: "You need to belong to an organization first." | Create an organization or accept an invitation, then start again. See Accounts. |
| Browser sign-in shows Your role allows less on Build or Full | Your role limits the level. Ask an owner for the member role in Members, or connect at Read. |
| "This agent's access was revoked or has expired. Connect it again." | The agent was revoked, its token ran out, a browser sign-in went 30 days without use, or you were removed from the organization or no longer have access to it. Connect again with browser sign-in, or create a new token. |
For every other message, see When something goes wrong.
Next steps
- Connect your app: the setup for Claude Code, Cursor, Codex, VS Code, Claude and ChatGPT, with a browser sign-in or a token.
- Tools: what an agent can do at each access level, how spending works, and how files get in.
- Access and tokens: access levels, personal tokens, revoking and limits.
Need a hand? Visit troubleshooting