Connect Claude to your flows over MCP

What it is

The MCP server lives at /api/mcp. It speaks stateless JSON-RPC over Streamable HTTP: every request carries a Bearer token, nothing is kept between calls, and there is no SSE stream to hold open. Any client that supports the HTTP transport can use it: Claude Code, Claude Desktop, or a script of your own.

The server publishes 11 tools. Seven work on documents: list_documents, read_document, search_documents, write_document, create_document, rename_document and share_document. Two work on history: list_versions and restore_version. audit_document runs the parse and audit checks without writing anything. get_syntax_guide returns the reading legend, and the server's own instructions tell the model to read it before writing.

How it works

A token is minted in the app and stored only as a SHA-256 hash; the raw value appears once, in the response that created it. On every call the server hashes the header value, looks it up, stamps last_used_at and resolves the owner. A suspended account resolves to nothing, so suspension cuts off API access as well.

read_document returns the document text with the legend appended, so the model gets the rules next to the content. write_document replaces the full text; for processes the response reports parse errors as 0-based line numbers and lists audit findings, so the model can fix and write again. Every write is followed by a version snapshot attempt, and restore_version attempts the same snapshot of the current text before replacing it, so a restore is usually reversible too. share_document returns the read-only /s/ link and the /embed/ URL, creating the link if none exists yet.

Why it matters

Your flows stop being a place you visit and become something your tools can read. Ask Claude Code to search every process for a term, audit the onboarding flow, or draft a refund process and hand back the share link, all without opening the editor. Because the documents are text and the server returns parse errors and audit findings on every write, the model corrects itself instead of leaving broken documents behind.

Limits

A token grants read and write access to your documents, so treat it like a password. Each account holds at most 5 tokens, and token creation is capped at 10 per hour. The endpoint allows 600 calls per hour per account and 20 messages per batch. Document text is cut at 100,000 characters and names at 200. search_documents returns the first 60 hits.

audit_document covers processes only; for a sequence diagram it reports that no audit exists yet. create_document shares the app's 60-per-hour creation bucket and, once billing is switched on, the free plan's cap of 10 processes and 10 sequence diagrams. rename_document over MCP changes only the name: .go and .do references are rewritten only when you rename inside the app. The server lists the documents your account created; a teammate's documents are not reachable through your token.

How to use it

1. 1
   Open Claude & MCP

   Click your avatar (the initial at the top right) and choose "Claude & MCP" from the menu. The modal lists your active tokens with their creation and last-used dates.
2. 2
   Create a token

   Click "Create a token". A green panel shows the token once with the note "Token created — copy it now, it will not be shown again." Press "Copy". Closing the modal clears it from the screen.
3. 3
   Add it to Claude Code

   Run the command the panel prints: claude mcp add --transport http runic https://runicflow.ai/api/mcp --header "Authorization: Bearer <token>". The server appears as "runic" in your MCP list.
4. 4
   Or connect Claude Desktop

   Add a custom MCP server with transport HTTP (streamable), endpoint https://runicflow.ai/api/mcp and the header Authorization: Bearer <token>. The same settings work for any MCP client.
5. 5
   Let Claude read the legend

   The server's initialize response asks the model to call get_syntax_guide before writing. Ask it to list your documents first, then work from names or ids.
6. 6
   Revoke when done

   Back in "Claude & MCP", click "Revoke" on any token row. The next call with that token gets a 401.

Commands & shortcuts

Tips

• Create one token per client so you can revoke one without touching the others. The modal names every token Claude; tell them apart by the created and last-used dates in the list.
• Documents resolve by id or exact name. Ask Claude to call list_documents first so it works from real names.
• write_document is a full replace. Ask for a read first, then a targeted rewrite, and check the parse and audit report that comes back.
• Use audit_document to review without changing anything, then write_document only once the findings are clear.
• restore_version tries to snapshot the current text before it replaces anything, under the same ten-minute throttle as any other write, so check list_versions first if the last snapshot is recent.

Questions