Skip to main content
The Bland MCP server accepts two credentials: your organization’s API key, or an OAuth 2.1 access token the client gets when you sign in with Bland. With sign-in there’s no key to create or paste. The client opens Bland in your browser, you sign in and approve what the client can do, and the client acts as you in the organization you chose.
Clients that already send an API key keep working as before. Nothing on this page changes them.

​
Who it’s for

  • ChatGPT users. ChatGPT (apps and connectors, in developer mode) connects to Bland by signing in today.
  • Client developers whose client identifies itself with a client ID metadata document. See For client developers.
Clients that only register dynamically (RFC 7591), including Claude, Claude Code, Cursor, and MCP Inspector, can’t sign in yet, because dynamic client registration isn’t available on Bland’s authorization server. Keep using an API key with them. Sign-in will apply to them once dynamic client registration is available.

​
What you see when you connect

1

Add the Bland server in your client

Give the client the server URL, https://api.bland.ai/v1/mcp. The client discovers Bland’s authorization server on its own and opens Bland in your browser.
2

Sign in

Sign in at v2.app.bland.ai with your phone number, Google, or SSO.
3

Choose an organization

If you belong to more than one eligible organization, pick the one the client acts in. The connection is scoped to that organization only. Organizations where your role is viewer, or where your account is banned, aren’t offered.
4

Review what the client gets

The consent screen lists the permissions the client asked for. To make a read-only connection, uncheck Create and change your agents. The full list is in Scopes and consent.
5

Use the tools

The client is connected. Every tool runs as you, in the organization you chose, with the permissions of your role there.
Tools declare what they need: Read tools need mcp:read, Write and Destructive tools need mcp:write. The tool reference shows each tool’s access level. If a tool needs a scope the connection wasn’t granted, for example a write tool on a read-only connection, the server answers 403 with a WWW-Authenticate challenge that lists the required scopes. Your client then asks you to grant them. The consent screen pre-checks what you already granted, so you approve only the addition.

​
Tokens and disconnecting

  • Access tokens are JWTs. Each one lasts 1 hour and is scoped to one organization.
  • If you granted Stay connected without signing in again, the client holds a refresh token and renews its access on its own.
  • To disconnect, remove or disconnect the Bland app in your client. The client revokes its tokens at https://api.bland.ai/authorization/oauth2/revoke.
  • A connected-apps page in the dashboard, for reviewing and revoking connections, is planned.
Over a sign-in connection, the purchase tools (buy_credits, buy_phone_plan) aren’t offered. Buy credits and phone plans in the dashboard, or over an API key connection. Every other tool behaves as it does with an API key, scoped to the organization you chose and the permissions of your role in it.

​
Client support

​
For client developers

Bland’s authorization server implements OAuth 2.1 authorization code with PKCE, and the MCP server publishes the discovery documents an MCP client expects. The protected resource document names https://api.bland.ai/authorization as the authorization server and mcp:read and mcp:write as the supported scopes. Bearer tokens are accepted in the Authorization header only. What your client must do:
  • Use PKCE with S256. It’s required on every authorization request.
  • Identify itself with a client ID metadata document. Your client_id is an HTTPS URL that serves your client’s metadata JSON. The server metadata advertises this with client_id_metadata_document_supported: true. Dynamic client registration (RFC 7591) isn’t available yet.
  • Send resource=https://api.bland.ai/v1/mcp (RFC 8707) in authorization and token requests. The token is only valid on /v1/mcp, not on the REST API.
  • Request the scopes it needs. mcp:read for read tools, mcp:write for write tools, plus openid, email, and offline_access as needed.
How the server responds:
  • No credential: 401 with WWW-Authenticate: Bearer realm="bland", resource_metadata="https://api.bland.ai/.well-known/oauth-protected-resource/v1/mcp", scope="mcp:read mcp:write". Follow resource_metadata to start discovery.
  • Missing scope: 403 with a WWW-Authenticate challenge that lists the required scopes. Re-run authorization requesting them; the consent screen pre-checks what the user already granted.
  • Scopes per tool: tools/list returns each tool’s required scopes in securitySchemes, with readOnlyHint, destructiveHint, idempotentHint, and openWorldHint annotations, so you can request the right scopes up front.
The token endpoint and the UserInfo endpoint are rate limited to 60 requests a minute per IP.

​
FAQ

Yes. Uncheck Create and change your agents on the consent screen. The client gets mcp:read only. If it later calls a write tool, it asks you to grant mcp:write.
No. buy_credits and buy_phone_plan aren’t offered over OAuth. Buy in the dashboard, or over an API key connection.
No. OAuth tokens are only valid on /v1/mcp. Use an API key for the REST API.
The one you chose at sign-in. Each access token is scoped to that organization. To work in a different organization, make a new connection and choose it.
No. A client that sends Authorization: Bearer org_... never sees the sign-in flow.
A connected-apps page in the dashboard is planned. Until then, disconnect from the client itself, which revokes its tokens.

​
Next steps

MCP clients

API key setup for Claude Code, Cursor, Codex, VS Code, and other clients.

Tool reference

Every tool the server exposes, with its access level.
Docs for agents: llms.txt