Using Weflow with Claude, OpenAI & Co via MCP

Connecting Weflow to Claude (and other AI assistants) via MCP

The Model Context Protocol (MCP) is an open standard for connecting AI assistants like Claude, Cursor, and Claude Code to external data. With MCP, your team can ask things like "summarize my last three calls with Acme" or "which deals had pricing objections this week" directly inside their AI assistant, and have it pull real data from Weflow on demand.

Important: Weflow does not ship an official MCP connector, and there is no hosted Weflow MCP server. This guide walks through two simple ways to bridge the Weflow Public API to an MCP client yourself, using tools that already exist in the open-source and commercial ecosystem.

Both options work because Weflow publishes a full OpenAPI 3.0 specification at https://api.getweflow.app/api/v1/docs. Any tool that can turn an OpenAPI spec into an MCP server will work with Weflow out of the box.

Before you start

You'll need:

A Weflow workspace with API access enabled, plus an API key. See Using the Weflow Public API for how to generate one.
An MCP-compatible client (Claude Desktop, Claude.ai, Claude Code, Cursor, and most modern AI IDEs all qualify).

Option 1: Run a local MCP server with one command

This is the fastest path for individual users. A community tool called openapi-mcp-generator reads our OpenAPI spec and produces a ready-to-run MCP server. You don't have to write any code.

Step 1. Make sure Node.js 20 or later is installed.

Step 2. Generate the server:

bash

npx openapi-mcp-generator \   --input https://api.getweflow.app/api/v1/openapi.json \   --output ./weflow-mcp \   --transport stdio

Step 3. Add it to Claude Desktop. Open your config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

Add a mcpServers entry:

json

{   "mcpServers": {     "weflow": {       "command": "node",       "args": ["/absolute/path/to/weflow-mcp/src/index.js"],       "env": {         "BEARER_TOKEN": "YOUR_WEFLOW_API_KEY"       }     }   } }

Step 4. Restart Claude Desktop. You'll see the Weflow tools appear in the input area, and Claude will use them automatically when your prompt calls for Weflow data.

Other generators in the same category that work the same way: openapi-to-mcp, openapi-mcpserver-generator. Pick whichever one you prefer.

Option 2: Use a hosted MCP generator service

If you don't want anything running on your laptop, a managed service can host the MCP server for you. You upload (or link to) Weflow's OpenAPI spec, configure authentication, and they give you back a public MCP server URL. You then add that URL to Claude as a custom connector.

Common services that offer this today:

Stainless (stainless.com)
Speakeasy Gram (speakeasy.com)
DigitalAPI (digitalapi.ai)

After you've set up the hosted server, add it to Claude:

In Claude.ai, go to Settings > Connectors
Click Add custom connector
Paste the MCP server URL from your hosted provider
Authenticate with your Weflow API key

Once connected, the Weflow tools are available in any conversation. The same connector also works in Claude Desktop, Claude Code, and the Claude mobile apps without further setup.

Tradeoff to be aware of: with a hosted generator, your OpenAPI spec and (depending on the service) your API key sit on a third party's infrastructure. Review their data handling and terms before pointing a production workspace at them.

What you can do once it's connected

A few example prompts that work well with the Weflow tools exposed by either option:

"List my Weflow recordings from the last two weeks with Acme and summarize the main objections."
"Pull the transcript from recording {uid} and draft a follow-up email covering the next steps we agreed to."
"Across last week's discovery calls, which competitors came up and in what context?"
"Score this call against our discovery template and tell me what I missed."
"Compare talk ratios across my team's calls this month."

Claude figures out which Weflow endpoints to call, chains them together, and works with the responses directly. You don't need to know the API yourself.

Things to keep in mind

The Weflow API key is workspace-scoped. Anyone with access to the MCP configuration has read access to every recording in your workspace. For team-wide deployments, host behind your own access layer.
Don't expose the /video endpoint to Claude. Pre-signed S3 URLs in an LLM context can be inadvertently shared, and they expire in an hour anyway.
Transcripts can be large. A 60-minute call can run to tens of thousands of tokens. If you hit context limits, ask Claude to use the summary endpoint first and only fall back to the full transcript when needed.
Rate limits apply. If Claude makes many sequential calls (e.g., looping over 50 recordings), you may see a 429. Most generated servers can be configured with retry/backoff.
You're responsible for the server. Weflow Support can help with API issues but cannot debug your MCP server or your AI client config.

Need help?

For Weflow API issues, email support@getweflow.com or use the in-app chat. For MCP-specific questions, the official MCP documentation is the best starting point.

Weflow Public API