Skip to main content
Tella offers an MCP (Model Context Protocol) server that lets you connect AI assistants like Claude directly to your Tella workspace.

​
What can you do with the Tella MCP server?

Once connected, AI assistants can:
  • List and search your Tella videos
  • Access video metadata, transcripts, and summaries
  • Manage video tags — list, create, update, delete, and assign tags to videos
  • Help you organize and manage your video library

​
Getting started

For full setup instructions, configuration options, and available commands, see our MCP documentation: tella.com/docs/mcp-server

​
Troubleshooting authentication

Connecting to the Tella MCP server runs an OAuth sign-in flow in your browser. When that flow gets stuck — the login never finishes, the AI client keeps reconnecting, or it worked before and suddenly stopped — it’s almost always one of the two issues below.

​
OAuth / callback loop

If clients like Claude Desktop connect through mcp-remote, authorizing opens a browser tab that hands a code back to a local callback. The connection hangs in a loop when that callback can’t complete. Try these in order:
1

Finish the sign-in in your browser

When the AI client starts, a browser tab should open asking you to authorize Tella. Complete that sign-in fully and wait for the “you can close this tab” confirmation before switching back. If no tab opened, your client may have suppressed it — check the client’s logs for an authorization URL and open it manually.
2

Make sure the local callback isn't blocked

The browser has to redirect back to a local callback URL on your machine. A VPN, proxy, or another process already using the callback port can break the handoff, leaving the client stuck reconnecting. Quit any other AI client that might be running the same MCP connection, then try again.
3

Fully restart the AI client

Quit the client completely (not just close the window) and reopen it so it starts a fresh connection rather than reusing a half-open one.
4

Authenticate once from a terminal (recommended for Codex)

Some clients restart an MCP server that isn’t ready within a short startup timeout — Codex allows about 10 seconds by default, less than a browser sign-in takes — and every restart opens another sign-in tab. Authenticate once outside the client instead: run npx mcp-remote https://api.tella.com/mcp in a terminal, complete the browser sign-in, wait for the confirmation page, then stop the command with Ctrl+C. The credentials are cached in ~/.mcp-auth, so the next time the client starts the connection it reuses them without opening a browser. For Codex you can also raise the timeout by adding startup_timeout_sec = 30 to the server’s section in ~/.codex/config.toml.

​
Reset stale credentials

If you used to connect fine and now you’re stuck — for example after changing your password, switching workspaces, or a token expiring — the cached credentials on your machine are likely stale and need to be cleared so the next connection re-runs the full sign-in. The MCP credential cache lives in your home directory at ~/.mcp-auth. Delete that folder — for example, run rm -rf ~/.mcp-auth in a terminal, or remove it in your file manager — to force a clean re-authentication on the next connection.
~/.mcp-auth is in your home directory (the ~ matters). It is not /.mcp-auth at the filesystem root.
After clearing the cache, fully restart your AI client and complete the browser sign-in again.

​
What is MCP?

Model Context Protocol is an open standard that allows AI assistants to securely connect to external tools and data sources. It was created by Anthropic and is supported by Claude and other AI applications.