StaticOwl in Claude Desktop

Get a free trial key → · Full MCP reference

What this gives you: the StaticOwl CMS as a first-class MCP server. Claude can list your sites, design content types, create and edit pages, manage releases, trigger builds, run the AI workflow — every meaningful CMS operation, exposed as a tool. You ask in chat, Claude does the work, the changes show up in your live site.

Cursor, Continue, Zed, and every other MCP-capable client work too. The endpoint is identical; the install snippet just lives in a different config file.

Step 1 — Request a trial key (no card)

Visit app.staticowl.com/signup-key.html, drop your email + a one-line "what are you building," submit. We email a verification link; click it, the page shows your so_* trial key + a copy-paste config block.

30 days · auto-provisions a real sandbox site with its own graph + seeded content
Full MCP access against your sandbox — schema, content, builds, the works
Free · no card, no credit checks, no chasing

Step 2 — Drop this into Claude Desktop

Open your Claude Desktop config:

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

And paste this (the verified-key page shows it with your real key already filled in):

{
  "mcpServers": {
    "staticowl": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://app.staticowl.com/api/mcp",
        "--header", "Authorization:Bearer YOUR_KEY_HERE"
      ]
    }
  }
}

Restart Claude Desktop. 38 StaticOwl tools show up under the integration picker.

Why mcp-remote? Claude Desktop is currently stdio-only. mcp-remote bridges to our HTTP server. When Claude Desktop ships native Streamable HTTP support (in flight) you'll drop the bridge.

Step 3 — Three prompts to try

"Show me my sandbox"

Use the StaticOwl MCP to list my sites. There should be one called "Sandbox" — show me its details (graph name, environments, theme) and tell me what content types it was seeded with. Then list the sample pages.

"Create your first content type"

In my Sandbox site, design a content type called "Recipe" with fields for title (string, required), ingredients (array of strings), instructions (richtext), prepMinutes (number), and a slug. Create it through your tools, then create three example recipes I can use as test data.

"Plan a docs site end-to-end"

I want to extend my Sandbox to be a documentation site for an open-source library. Walk me through every change you'd make — new content types (Doc, Section, ApiReference), sample content, and the template you'd wire up. Make the changes through your tools and tell me when each step succeeds.

What the 38 tools cover

Area	Tools	What you can do from chat
Sites	5	List, get, create, update, set-default
Content	10	List, get, create, update, delete, publish, drafts, search, comment, transitions
Schema	6	Types, queries, scripts — list / get / create / update
Templates	4	List, get, create, update
Builds + releases	6	Trigger, history, impact, rollback, release-bundle, deploy
Introspection	4	Views, routes, site structure, search
System	3	Health, schema export, status

For the full schema of every tool, see the MCP reference or call tools/list on the endpoint.

Why MCP-native matters

Most "AI CMSes" are AI bolt-ons: a chat sidebar that calls one or two endpoints, with the rest of the product still requiring a human in the admin UI. StaticOwl is the inverse — every meaningful operation is exposed as an MCP tool from day one, so a capable model can drive the whole product:

Drive vs. assist: Claude doesn't suggest changes for you to click through. It makes them, then reports back.
Same authority as a human admin: Whatever permissions your key has, the model has. Trial keys are sandbox-scoped; full-account keys aren't.
Audit-grade backbone: every change goes through the bitemporal graph engine, so "what did the AI do at 2:14pm last Tuesday" is a query, not a forensics project.
Cursor / Zed / Continue support: same endpoint. Different config file.

Curl it directly (no client needed)

The whole MCP server is HTTP-addressable. You can call it from anywhere:

# Health check
curl https://app.staticowl.com/api/health

# Who am I?
curl -H "Authorization: Bearer $KEY" https://app.staticowl.com/api/access/me

# List every MCP tool the server exposes
curl -X POST https://app.staticowl.com/api/mcp \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

# Call a tool (here: list sites)
curl -X POST https://app.staticowl.com/api/mcp \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"sites_list","arguments":{}}}'

FAQ

Does this work with Cursor, Zed, Continue?

Yes. Any MCP-capable client. The endpoint is the same; the install snippet lives in their respective config files. We'll publish per-client guides as they get used in the wild — meanwhile, mcp-remote works as the bridge for any stdio-only client.

What happens at day 30?

Your sandbox + trial key expire. If you've subscribed to a paid plan before then, the sandbox migrates into your real account. If not, the sandbox is dropped. Subscribing later is one click — your account stays so you can come back to it.

Is it really MCP-native? Or a bolt-on?

Native. The MCP server is the same HTTP surface as the CMS API — every tool is an Express route under the hood. Trial users get the full surface (read + sandbox-scoped writes); paid users get unrestricted access scoped to their account. No "AI features" tier — MCP is the product.

Where can I see the live tool catalog?

Hit tools/list on https://app.staticowl.com/api/mcp with a valid key, or read the human-friendly version at staticowl.com/docs/mcp.

Get started

Free 30-day trial. Email verification only. No card. No install on your machine. Get my trial key →

Questions? Email support@staticowl.com.