Bland + MCP

Build and test Bland voice agents from Claude Code, Cursor, or Codex through Bland's hosted MCP server: place calls, ship agents, and read transcripts.

Book a call Read the docs

What Bland does inside MCP.

Build on the stack Bland runs itself.

Agents you build and call through MCP run on Bland's self-hosted voice stack, respond in 200 milliseconds, and sit on a SOC 2 Type II and HIPAA ready platform.

Hosted, nothing to run.

Point any MCP client at api.bland.ai/v1/mcp over streamable HTTP. Bland hosts the server and it is stateless, so there are no sessions to manage.

Calls from a prompt.

create_call, wait_for_call, and get_call_log let your agent place a call, wait for it to finish, and read the transcript, recording, and summary.

Build and ship voice agents.

Validate a pathway graph before saving, and create, version, publish, and promote Bland Agents without leaving your editor.

Analytics and evals.

query_analytics answers questions about your calls with grouping, filters, and time ranges. Eval tools list judges and run evals over batches of calls.

Side effects need a yes.

Every tool is marked read-only or destructive, and mutating tools require an explicit confirmation argument, so a well-behaved client asks before it places a real call.

How it deploys.

Pick your client.
Claude Code, Cursor, OpenAI Codex, VS Code, Claude Desktop, Windsurf, and Gemini CLI each have a setup guide in Bland's docs. Any other MCP client works too.
Add the server.
Point the client at https://api.bland.ai/v1/mcp, or install the Norm plugin to get the server plus Bland Agents skills.
Authenticate.
Sign in with Bland where the client supports it, or set your Bland API key in the Authorization header.
Try a safe first task.
Ask your agent to list your pathways, then to place a test call to your own number and read back the transcript.

Our FDE team scopes the integration, maps Bland to your stack, and stays on through go-live. Enterprise deployments go live in 30 days.

Talk to an FDE

Drop Bland into your editor.

MCP server runs in Claude Code, Cursor, and any MCP-aware editor. Or hit the REST API directly.

1
2
3
4
5
6
7
8
curl -X POST https://api.bland.ai/v1/calls \
  -H "Authorization: $BLAND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "phone_number": "+14155552120",
    "task": "Qualify refi lead",
    "voice": "june"
  }'

Visit Trust and Security Portal

Frequently asked questions.

Does Bland have an MCP server?

Yes. Bland hosts a remote MCP server at api.bland.ai/v1/mcp. Coding agents and compatible personal AI agents use it to place and inspect calls, send texts, build pathways, manage agents, query analytics, run evals, and search the Bland docs.

Which MCP clients work with Bland?

Claude Code, Cursor, OpenAI Codex, VS Code with GitHub Copilot, Claude Desktop, Windsurf, and Gemini CLI have step-by-step guides. Any other client can connect over streamable HTTP, a stdio bridge, or raw JSON-RPC.

How does the Bland MCP server authenticate?

Sign in with Bland over OAuth where the client supports it, today ChatGPT, Claude, and Claude Code, or send your Bland API key as a bearer token, which works with every client. Either way, one organization scopes every tool.

Can an AI coding agent place real phone calls through MCP?

Yes, with confirmation. create_call places an outbound call and is marked destructive, so well-behaved clients ask first, and mutating tools also require an explicit confirmation argument. The server allows 120 requests per minute per organization.

What is the Norm plugin?

Norm, installed as norm@bland, bundles the Bland MCP server with Bland Agents skills and /norm commands for coding agents like Claude Code, so you can build, migrate, validate, and test Bland Agents from your editor.

Can a personal AI agent get its own phone number?

Yes. Assistants such as OpenClaw, Poke, and Meta Muse can connect to Bland through MCP or REST and get a dedicated number on the Agent Phone Plan, with unlimited calling and texting to the US and Canada.