Help center
Sign up
Agents

Connect table to AI clients with MCP

Use table's MCP server to let compatible AI clients work with table through approved, narrow tools.

What MCP does

MCP, or Model Context Protocol, is a standard way for AI clients to connect to tools and data from other apps. table's MCP server lets compatible AI clients request access to your table account and use approved table tools.

The MCP server is hosted separately from the main app and uses its own authorization flow. This keeps the MCP boundary explicit.

When to use MCP

Use MCP if you want to:

  • Save relationship context from an AI client into table
  • Create or update contacts from explicit contact data
  • Let an assistant help maintain your CRM without giving it broad database access
  • Build workflows that start outside table but end in your table account
Do not use MCP for sensitive workflows unless you trust the client you are connecting.

How authorization works

When an MCP client requests access, table shows an authorization screen. The screen identifies the client and asks you to approve or cancel.

If you are not signed in, table asks you to sign in before authorization can continue.

Approving a client allows that MCP client to use the table MCP server according to the tools and access currently available. You should only approve clients you recognize.

Connect an MCP client

The exact setup depends on the client. In general:

  1. 1Open your MCP-compatible client.
  2. 2Add a remote MCP server.
  3. 3Enter the table MCP server URL.
  4. 4Start the authorization flow from the client.
  5. 5Sign in to table if prompted.
  6. 6Review the client name and redirect details.
  7. 7Select Approve.
  8. 8Return to the client and test the connection.
Opening the authorization URL manually may be missing required request details. Start from your MCP client.

Available tools

The current MCP surface is intentionally narrow. table MCP tools should be product-shaped, not raw database access.

The first tool is focused on saving one contact from explicit data, including fields like name, email, phone, organization, job title, location, birthday, and status.

As table MCP grows, tools should remain narrow, permissioned, and understandable.

Security model

  • MCP clients must authenticate.
  • The server derives the user from the authenticated token, not from tool input.
  • Tools validate inputs before running.
  • Write tools use table's server-owned product logic.
  • Tools should not expose raw database rows.
  • Errors should be stable and bounded.
  • Logs should avoid personal data such as names, emails, phone numbers, organizations, prompts, tokens, and provider payloads.

Troubleshooting

  • Invalid authorization request: start the authorization flow from your MCP client.
  • Unknown client: check that the client registration is valid and that the redirect URL is expected.
  • Approved client cannot call tools: disconnect and reconnect the MCP server in the client.
  • Unrecognized client: cancel authorization and do not approve the request.

FAQ

No. table AI is the assistant inside table. MCP is a way for external AI clients to connect to table through approved tools.

No. MCP access depends on the tools table exposes. The intended design is narrow tools with bounded inputs and outputs, not raw database access.

Yes. Revoke access for clients you no longer use. If a client does not look familiar, cancel the authorization flow and contact support.