Every connection gets both read and write access. There is no read-only credential, so connect an agent only if you are comfortable with it creating and editing documents on your behalf.
Sign in from the client
This is the path most MCP clients take. Add the server to the client’s MCP configuration:mcp-config.json
Connect Claude Code. If you belong to one workspace, it names that workspace too:
Claude Code wants to connect to the Acme workspace. It will be able to read, create, edit, move and delete documents and folders in this workspace only.If you belong to more than one, the screen reads “Claude Code wants to connect to a Tatara workspace. Choose one below.” and lists your workspaces by name. Tatara does not pick one for you, and Connect stays disabled until you choose. Choose Connect to approve. The page shows
Connected to Tatara and takes you back to your app. If nothing happens after a few seconds, select Click here if nothing happened. Choose Cancel to refuse instead, and the page confirms the connection was cancelled and that you can close the tab.
From then on the client keeps its own access current, so you do not have to sign in again for every session. A client that goes 30 days without using its connection has to sign in again.
Browser sign-in works with clients that hand control back to an address on your own machine, which covers desktop and command-line clients. Of the clients that run on their provider’s servers instead, Claude on the web is the one Tatara accepts. If your client cannot complete the browser round trip, use an access token instead.
Client by client
Claude Code and Claude Desktop take the entry above in their MCP configuration. Claude on the web (claude.ai) takes Tatara as a custom connector. In your connector settings, add a custom connector pointing athttps://www.tatara.app/api/mcp, then sign in to Tatara and approve the same consent screen. On a Claude Team or Enterprise plan, an owner adds the connector once and each person then connects it for themselves.
Gemini CLI reads ~/.gemini/settings.json, or a project’s .gemini/settings.json if you want the connection scoped to one repository. Its entry gives the address as httpUrl:
settings.json
headers object carrying an access token, as in Use an access token.
One entry per workspace
A connection stays locked to the workspace you chose when you approved it. Every read and write goes there, and the agent has no way to switch. To reach a second workspace from the same client, add a second entry with a different name and the same address:mcp-config.json
Use an access token
An access token is a long-lived credential for an agent that cannot sign in through a browser, or that runs unattended. It reaches the workspace of the account it was created in. It has no expiry date and works until someone revokes it. Only an account owner can create or revoke a token, and Access tokens appears in settings only for owners. A token belongs to the account rather than to the person who created it.1
Create the token
In Tatara, open Settings → Access tokens and choose Create token. A new row opens in the list. Type a name that identifies the agent using it, then press Enter.
2
Copy it immediately
The token appears in that row, once. Choose Copy, then Done. Tatara does not store a readable copy, so a lost token has to be replaced.
3
Send it on every call
Configure the client to send the token as a bearer credential:
mcp-config.json
Manage connected agents
Settings → Agent access lists, under Connected apps, the clients you connected to this workspace through the browser. Each row shows the date the client connected and when it last called, ornever used if it has not called yet. The list is yours alone: you see the clients you connected, and each member sees their own. The Agents count in the status line at the bottom of the app is the number of clients on this list.
To end a connection, choose Disconnect on its row and confirm. The client is cut off on its very next request, the same as a revoked access token. To use it again, connect it from the client.
Agents that use an access token are not listed here. Manage them under Settings → Access tokens.
Provenance
Every document an agent writes records which connection wrote it, in the document’ssource property. For an access token that is the token’s name, so name each token after the agent that uses it. For a client signed in through the browser it is the client’s identifier. The agent cannot set or change it.
If the connection fails
The page says the sign-in request has expired
The page says the sign-in request has expired
A consent request lasts five minutes. Start the connection again from your client and approve it without leaving the tab idle.
The consent screen says to finish setup first
The consent screen says to finish setup first
You need a workspace before an agent can connect. Choose Finish setup, then restart the connection from your client.
The consent screen says something went wrong
The consent screen says something went wrong
Go back to your client and start the connection again.
The agent says its token is not recognized or its connection has ended
The agent says its token is not recognized or its connection has ended
The token was revoked or mistyped. Create a new one under Settings → Access tokens and update the client’s configuration. If the agent connected through the browser instead, its connection has ended: it was disconnected under Settings → Agent access, approved again for another workspace, or the person who connected it lost their membership. Connect it again from the client.
Next steps
The MCP server
What your agent can do once it is connected.
Agent skills
Install the skills that teach your agent to write correctly.