Docsie Video-to-Docs MCP Server

Remote MCP server for Anthropic's Connectors Directory. Converts videos into structured documentation using Dokuta AI.

What It Does

Users on Claude can:

1. Submit a video URL (YouTube, Loom, Vimeo, direct MP4, etc.)

2. Choose a doc type (user guide, SOP, product docs, policy, blog)

3. Get structured markdown documentation back

Free tier: Videos up to 5 minutes are processed free. Longer videos require a Docsie account with AI credits.

Architecture

Claude → POST https://mcp.docsie.io/mcp (Bearer token)
              │
              ├─→ Docsie /o2/ (OAuth2 token validation)
              ├─→ Dokuta API (video analysis)
              └─→ Docsie /api/internal/mcp/ (credit deduction)

• Transport: Streamable HTTP (MCP spec)
• Auth: OAuth2 Authorization Code + PKCE via Docsie's `/o2/` provider
• Runtime: Python 3.12 + FastMCP + Starlette + uvicorn

MCP Tools

Local Development

cd mcp-server

# Install dependencies
pip install -e .

# Configure environment
cp .env.example .env
# Edit .env with your Dokuta API key and Docsie internal URL

# Run
uvicorn app.main:app --reload --port 8000

Test endpoints

# Health check
curl http://localhost:8000/health

# OAuth metadata
curl http://localhost:8000/.well-known/oauth-authorization-server

# MCP tool listing (via MCP protocol)
curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":"1","method":"tools/list","params":{}}'

Environment Variables

Deployment (Kubernetes)

Docker images are built and pushed automatically via GitHub Actions on push to `main`.

# Deploy with Helm (secrets managed via Doppler)
helm upgrade --install mcp-staging helm/docsie-mcp/ \
  -n mcp \
  --set image.tag=v2026.04.06.01 \
  --set doppler.token=$DOPPLER_SERVICE_TOKEN

Django Side Setup

1. Add internal API key to Django settings

In your `.env` or environment:

MCP_INTERNAL_API_KEY=your-shared-secret-here

2. Register OAuth2 Application

In Django admin at `/o2/applications/register/`:

• Name: Claude MCP Connector
• Client type: Public
• Authorization grant type: Authorization code
• Redirect URIs: (provided by Anthropic during directory registration)
• Skip authorization: Yes

3. Internal endpoints (already wired)

• `GET /api/internal/mcp/user-context/` — token validation + user info
• `POST /api/internal/mcp/deduct-credits/` — credit deduction

Anthropic Connectors Directory Submission

To submit to the directory, you'll need:

1. MCP Server URL: `https://mcp.docsie.io/mcp`

2. OAuth2 metadata URL: `https://mcp.docsie.io/.well-known/oauth-authorization-server`

3. Company name: Docsie Inc.

4. Description: "Convert videos into structured documentation using AI. Submit a video URL and get back a user guide, SOP, product docs, or blog post."

5. Logo: Docsie logo (SVG/PNG)

6. Privacy policy: https://www.docsie.io/privacy-policy/

7. Terms of service: https://www.docsie.io/terms-of-service/

Submit at: https://console.anthropic.com/dashboard/connectors