MCP server - CaptureFlow API

CaptureFlow’s remote MCP server lets an AI assistant work with your CaptureFlow content: read your ideas, inputs and posts, create ideas and drafts, and propose posts for your approval.


Server URL	`https://mcp.captureflow.ai/mcp`
Read-only server URL	`https://mcp.captureflow.ai/mcp/readonly`
Transport	Streamable HTTP
Sign-in	OAuth 2.1 (the assistant opens CaptureFlow’s sign-in), or an API key header

Nothing is ever published without your approval. An assistant can propose a post; you accept or reject it in CaptureFlow.

Connecting

Add the server URL in your assistant (see the page for your assistant).
Sign in to CaptureFlow when the assistant asks.
Choose the workspace and what the assistant may do: Read only, Read and create drafts or Full access. You can narrow what it asked for.

Your connected assistants are listed in Settings > API & MCP > Connected apps, where you can disconnect them at any time. Read-only URL: an assistant connected to /mcp/readonly can read, never change anything. Use it when you only want answers about your content.

Tools

Each tool matches an API operation and needs the same scope.

Tool	Does	Scope
`whoami`, `list_workspaces`	Who you are; your workspaces
`list_ideas`, `get_idea`	Your content ideas	`ideas:read`
`create_idea`	Add an idea	`ideas:write`
`list_inputs`, `get_input`	Your inputs, with transcripts and processing state	`inputs:read`
`create_text_input`, `import_url_input`	Add text or a web page as an input	`inputs:write`
`create_input_upload`, `get_input_upload`, `refresh_input_upload_url`, `complete_input_upload`	Upload a file (needs an assistant that can PUT a file)	`inputs:write` / `inputs:read`
`list_posts`, `get_post`	Your posts	`posts:read`
`check_post_preflight`	Check a post against platform rules	`posts:read`
`create_post_draft`, `update_post_draft`	Write and edit drafts	`posts:write`
`propose_post_schedule`, `withdraw_post_proposal`	Propose a post for your approval, or take it back	`posts:schedule`
`list_social_accounts`	Your connected accounts	`accounts:read`
`get_strategy`	Your content strategy	`strategy:read`
`start_content_agent`, `get_agent_run`	Let the content agent write drafts	`agent:run`

Tools are annotated: reads are readOnlyHint, and only update_post_draft and withdraw_post_proposal are destructiveHint (they replace or take back something).

Good to know

Workspaces: when you belong to several, tools take a workspace_id; list_workspaces lists them. See Workspaces.
Safe retries: tools that change something take an optional idempotency_key, except refresh_input_upload_url and complete_input_upload, which are safe to repeat on their own (conventions).
More permissions: if a tool needs a scope you didn’t grant, the assistant asks you to approve it (or tells you to reconnect).
Errors come back as tool errors with the same code as the API, explained on the errors page.
Your data is used only to serve you, never to train AI models. The assistant receives only what its permissions allow.

​Connecting

​Tools

​Good to know

Connecting

Tools

Good to know