The DEPLOY connector · for Claude & ChatGPT

Use DEPLOY in Claude or ChatGPT

Connect once. Then just ask, right in the chat, to look up any robot or company, fact-check a claim before you publish, or order food across DEPLOY Ground. No new app to learn.

Add DEPLOY to your assistant

Add to Claude

Claude: one click on Desktop, or paste the URL below into claude.ai → Settings → Connectors → Add custom connector. ChatGPT (Plus, Pro, or Business): tap Add to ChatGPT to copy the URL, then in ChatGPT open Settings → Apps → turn on Developer Mode → Add custom MCP server, and paste it.

The first time your assistant uses a personal tool, like your watchlist or placing an order, it asks you to sign in to DEPLOY once. After that, it acts as you.

Other apps & manual setup (Cursor, Claude Code, config file)

Any MCP-capable client can use DEPLOY. Point it at the URL, or use the bridge command for clients that only accept a command:

URL: https://deploy.report/mcp
Bridge command: npx -y mcp-remote@latest https://deploy.report/mcp

Claude Desktop config (Settings, Developer, Edit Config), or Cursor (Settings, MCP, Add new server):

{
  "mcpServers": {
    "deploy": { "url": "https://deploy.report/mcp" }
  }
}

Older clients without remote-URL support use the bridge command above in place of the URL.

Try these prompts

Once connected, open a new conversation and just ask:

What robots does Waymo operate in San Francisco?
Is Physical Intelligence really valued at $11B?
Audit this paragraph before I publish: [paste your draft]
Find restaurants I can order from with my AI in Los Angeles.
Add Figure to my DEPLOY watchlist.
Compare RingConn Gen 2 and Happy Ring on FDA clearance.

Your assistant calls the DEPLOY tools automatically when they're relevant. You don't have to invoke them by name.

What you can do with DEPLOY connected

Find and order across DEPLOY Ground

Ask your assistant to find restaurants and businesses it can reach, check if they're open, and place an order on your behalf. Connect once and it works everywhere on Ground, with no separate app per restaurant. Personal actions ask you to sign in to DEPLOY the first time, then run as you.

Look up anything in the registry

Robots, manufacturers, AI brains, deployments, incidents, regulations. Your assistant pulls the canonical record on demand: sourced, dated, human-reviewed. No need to leave the conversation.

Fact-check claims as they cross your desk

Paste a line from an article, a press release, or a pitch deck. DEPLOY tells you whether it's verified, flags known aggregator errors (inflated valuations, FDA-status conflations, deployment overstating, founder-pedigree mix-ups, corporate-state errors), and surfaces the corrected version with sources you can cite.

Audit AI-generated drafts before publishing

Run a draft through DEPLOY. Every claim gets checked against the verified record. Conflicts are flagged inline with the honest framing for each one, plus a compliance score across the whole piece.

Compare products on what actually matters

Side-by-side comparisons that go beyond spec sheets: FDA-clearance posture, autonomy level (supervised vs fully autonomous), business model, and the architectural pattern each product represents. Your assistant surfaces the framework-load-bearing differences, not just the marketing ones.

Why this matters

AI assistants hit training-set walls fast on physical AI; the field moves quicker than training corpora catch up. DEPLOY is live, sourced, and human-reviewed: connecting your assistant means answers reflect what's actually verified in the world, not what the training data happened to capture.

DEPLOY also draws an explicit line between verified facts and manufacturer claims. When your assistant queries DEPLOY, that distinction surfaces in every response. Read the methodology for the full editorial framework.

Help + troubleshooting

If something doesn't work:

Confirm the server is reachable: https://deploy.report/mcp in a browser should show this page
If using the bridge config and seeing "npx not found", install Node.js first
If your client doesn't support MCP at all yet, use the REST API directly
For corrections or partnership questions, see cite

For developers

Direct API access:

Endpoint: POST https://deploy.report/mcp
Protocol: JSON-RPC 2.0 over HTTP. Stateless. Public read tools are anonymous; personal tools (order, watchlist, account) use OAuth 2.1 (see protected-resource metadata).

Test from a terminal:

curl -X POST https://deploy.report/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'