Flow2

What the flow2 MCP server does

The flow2 MCP server gives an AI assistant access to the Flow2 authoring workflow. It can create a presentation, change its metadata, inspect its current state, and manage the screens that make up the flow. Within each screen, the assistant can add, update, inspect, or remove blocks such as text, images, and buttons.

The server also supports presentation review and delivery tasks. An assistant can request a screenshot for an individual screen, render a complete flow as a vertical preview, and show a phone-framed preview for one screen. It can publish changes through the available Flow2 workflow and check account credits. Long-running operations can be monitored with job-listing and job-status tools.

How it works

Flow2 is exposed through a remote streamable-HTTP MCP endpoint at https://mcp.flow2.co/. A compatible client sends MCP requests to that endpoint, and the server applies those requests to the authenticated user’s Flow2 account. The integration uses OAuth 2.1 and Dynamic Client Registration: the client discovers the authorization metadata, registers itself, completes an authorization-code flow with PKCE, and retains a refresh token for later access.

Clients with MCP Apps support can display render-flow-preview and render-screen-preview directly in chat using a sandboxed iframe. The full-flow view can show all screens with a theme-accent indicator, while the screen view presents one screen in a phone frame. On clients without that extension, preview requests fall back to the cover screenshot. CTA buttons in individual-screen previews can act as click-through links where the inline preview is supported.

Setup and configuration

The flow2 MCP server does not require local environment variables in the documented setup. Connect a remote MCP client to https://mcp.flow2.co/ and complete the OAuth sign-in when prompted.

Claude Code can be configured with claude mcp add --transport http flow2 https://mcp.flow2.co/ --scope user, followed by /mcp to finish authorization. Claude.ai web and desktop users can add the URL as a custom connector. Cursor can add the endpoint under the flow2 entry in ~/.cursor/mcp.json. Other clients need support for remote streamable-HTTP MCP servers; OAuth is handled automatically when Dynamic Client Registration is supported.

Access is limited to Flow2 data belonging to the authenticated user. Request metadata such as tool name, timestamp, and status is logged for operations, while tool arguments and response bodies are not retained beyond the request lifecycle. Authentication tokens are encrypted at rest, and revoking access in Flow2 settings purges them.

Tools and capabilities

The flow2 MCP server exposes these tool groups:

• Flow management: create_flow, update_flow_meta, and get_flow.
• Screen management: add_screen, get_screen, and delete_screen.
• Block editing: add_block, patch_block, delete_block, and get_blocks_schema.
• Preview generation: get_preview, render-flow-preview, and render-screen-preview.
• Operation monitoring: list_jobs and check_job_status.
• Account information: check_credits.

Read-only tools are marked with MCP’s readOnlyHint annotation. Appropriate write tools use destructiveHint: false, allowing clients to reason about the operation’s safety. The available block types and their schemas can be retrieved before constructing block content, which helps an assistant create valid screen elements.

The repository is licensed under MIT. Support is available through the project’s GitHub issues page and the listed Flow2 email contact.