Connect Pendro to Claude Desktop (and every other AI agent) via MCP

Claude Desktop, Claude in Chrome, ChatGPT, Zapier MCP, and Cursor all speak Model Context Protocol — an open spec for connecting AI agents to external tools. Pendro ships an MCP server so any of those clients can drive your account conversationally. Ask Claude what projects you have, generate a new landing page from a prompt, draft a blog post, read your contact form inbox — all from a chat window.

Here's how to wire it up.

What you get

Seventeen tools, mapped 1:1 onto Pendro's existing REST API. The agent reads each tool's description, decides which to call, and chains them together to fulfil your request:

• Projects: list, get, create (from a template or AI prompt), update, delete.

• Blog posts: list, get, create, update, delete — per project.

• Form submissions: read your contact-form inbox.

• Newsletter subscribers: read your mailing list.

• AI: generate full projects from a brief, regenerate individual sections, read/write brand voice.

Two ways to connect

Pick the transport that fits your setup:

A) Remote HTTP — Custom Connectors dialog

Open Claude Desktop → Settings → Custom Connectors → Add custom connector. Paste:

https://app.pendro.co/api/v1/mcp?token=pendro_pat_YOUR_TOKEN_HERE

Skip the OAuth Client ID / Secret fields (Pendro doesn't expose an OAuth provider yet). Click Add. The connector attaches within a second — tool count appears in the connectors panel.

B) Local stdio — JSON config file

If you'd rather run a local process (lower latency, works offline, no token in URL), install the npm package and add it to your Claude Desktop config:

npm install -g @app/mcp
pendro-mcp doctor --api-key pendro_pat_YOUR_TOKEN

The doctor command pings the API + verifies your token without starting the MCP server — a sanity check before Claude Desktop tries to attach.

Then edit your Claude Desktop config at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) and add:

{
  "mcpServers": {
    "pendro": {
      "command": "npx",
      "args": ["-y", "@app/mcp", "serve"],
      "env": {
        "PENDRO_API_KEY": "pendro_pat_YOUR_TOKEN",
        "PENDRO_API_URL": "https://app.pendro.co"
      }
    }
  }
}

Restart Claude Desktop (Cmd+Q on macOS, not just close). The Pendro server appears in the bottom-left of the chat window.

What to ask Claude

Some prompts that actually work:

• "What projects do I have on Pendro?" → list_projects.

• "Show me the content of my Climbird project." → get_project.

• "Create a new project called Atlas Therapy using the wellness-spa template." → create_project.

• "Change Climbird's primary brand colour to deep purple." → get_project then update_project with a theme patch.

• "Draft a blog post for Climbird titled 'Why we built this' — about three paragraphs." → create_blog_post with status: draft.

• "Generate a landing page for a boutique pilates studio in Brooklyn focused on postnatal recovery." → generate_project_from_prompt.

• "Make the hero on Climbird more playful — less corporate." → regenerate_section with a tone hint.

How it stays safe

• Per-token scopes: mint a token with only the permissions you need.

• Rate-limited: 60 requests per token per minute, server-side.

• Token revocable: delete a token from /settings/api-tokens and the connector 401s immediately.

• No silent admin escalation: every server action runs the same auth + plan-gating checks as the dashboard UI.

• Audit log: admin-side audit log captures who-did-what; agent calls show up the same as a human dashboard click.

Beyond Claude Desktop

The same URL works with Zapier MCP (drop into your Zap's Custom Connector field), ChatGPT MCP (when OpenAI's MCP support lands), Cursor (Cursor has an MCP panel under Settings → Models), and any future MCP client. The stdio CLI works with Cursor and Continue.dev too.

Try it

Mint a token, paste the URL into Claude Desktop, and ask Claude what projects you have. The full setup takes about three minutes.