BioFlow MCP

npx add-mcp 'https://app.getbioflow.com/api/mcp'

Two minutes to your first tool call.

One remote endpoint

Streamable HTTP, no local install and no API keys to copy around:

https://app.getbioflow.com/api/mcp

Claude Code

Add BioFlow from your terminal, then sign in when prompted:

claude mcp add --transport http bioflow https://app.getbioflow.com/api/mcp

Every other client

Claude web and desktop (as a custom connector), ChatGPT, Copilot, MCP Inspector — the setup guide has the exact steps for each, plus the OAuth discovery endpoints if you are building your own client.

Read the setup guide →

13 tools, one scope each

Read pages, analytics, contacts and files; create and edit drafts; publish. Every tool is guarded by a single OAuth scope, and tokens are scoped to one workspace — an agent only ever sees the workspace you connected it to.

See the full tool and scope reference →

Built so an agent can't surprise you.

Consent & revocation

Every connection starts on a consent screen listing each requested scope in plain language. Revoke any app in Settings → Connected AI apps — its next request is refused, and the same screen shows a full audit trail of every tool call the app ever made.

Publishing is off by default

Block edits wait in the draft; title, description and theme edits apply to your live page without waiting for a publish. The publish-class tools are disabled per workspace until you flip an explicit settings toggle, and even then every publish is a two-step confirm — the first call only ever returns a preview.

Nothing gets clobbered

Draft writes are guarded by optimistic concurrency, so a stale edit is refused rather than overwriting newer work, and writes take an idempotency key so a retried call can't run twice.

Structured, actionable denials

Every refusal is a typed code an agent can act on — a missing scope, a disabled dangerous operation (with a deep link to the setting), a plan limit, a stale draft. No mystery errors, no silent no-ops. Full denial reference →

Agent questions, answered.

Is my data safe when I connect an AI agent?

Yes. Agents connect through OAuth 2.1 — you approve exactly which scopes an app gets on a consent screen, tokens are limited to one workspace, and every tool call is written to an audit trail you can review in Settings → Connected AI apps. You can revoke an app at any time, and its next request is refused.

What can an agent do without my approval?

Only what its granted scopes allow. With write access it can edit your blocks, which wait in the draft until you publish, and your page's title, description and theme, which apply to your live page without waiting for a publish. Publishing and scheduling are off by default per workspace, and even after you enable them, every publish is a two-step flow: the agent gets a preview and a short-lived confirm token first, and nothing commits until a second call with that token.

How do I disconnect an AI app?

Open Settings → Connected AI apps in your BioFlow dashboard and revoke the app. Its next request to your workspace is refused, and its refresh token is revoked, so it can't get new access.

Does MCP access cost extra?

No. The MCP server is included on every plan, including Free. Tool calls ride your plan's existing limits — an agent can't spend more than your plan allows, and over-limit calls get a structured QUOTA_EXCEEDED response instead of silent failure.

Which AI agents work with BioFlow?

Any MCP client that speaks Streamable HTTP with OAuth: Claude (web and desktop, as a custom connector), Claude Code, ChatGPT, Microsoft Copilot, MCP Inspector, and any agent framework with an MCP client. Registration is dynamic — no pre-shared API keys needed.

Setting it up: the MCP setup guide. More on BioFlow itself: the full FAQ, the integration directory, and pricing.