Connect an agent to ARC

MCP

Streamable HTTP. Discovery for both servers is at https://www.arcreport.ai/.well-known/mcp.json and the server card for the scan server is at /api/mcp/server-card.

Claude

Claude.ai and Claude Desktop: Settings → Connectors → Add custom connector. Paste https://www.arcreport.ai/api/mcp, then add https://www.arcreport.ai/api/ask if you want Index reads. Choose no authentication.

claude mcp add --transport http arc https://www.arcreport.ai/api/mcp
claude mcp add --transport http arc-ask https://www.arcreport.ai/api/ask

Older Claude Desktop builds can bridge with mcp-remote:

{
  "mcpServers": {
    "arc": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://www.arcreport.ai/api/mcp"]
    },
    "arc-ask": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://www.arcreport.ai/api/ask"]
    }
  }
}

ChatGPT

Settings → Apps. Where the account offers Developer mode, turn it on, add a custom MCP app, paste https://www.arcreport.ai/api/mcp, and choose no authentication. Menu names vary. OpenAI describes the steps in this help article.

Gemini

Gemini Spark can take a custom MCP URL. On a computer, open gemini.google.com → Settings & help → Connected Apps → Custom apps for Spark → Add a custom app, and paste https://www.arcreport.ai/api/mcp. Google's help says this is for Spark, in English, on a personal Google account with Keep Activity on, and that a server without dynamic client registration needs credentials under Advanced. ARC's public tools do not need a key. See Google's custom-apps article.

Cursor

Project file .cursor/mcp.json, or the global ~/.cursor/mcp.json:

{
  "mcpServers": {
    "arc": { "url": "https://www.arcreport.ai/api/mcp" },
    "arc-ask": { "url": "https://www.arcreport.ai/api/ask" }
  }
}

Generic agents

Any client that speaks MCP streamable HTTP can POST JSON-RPC to the server URL. initialize, tools/list, and the public tools need no token.

curl -s https://www.arcreport.ai/api/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

OpenAPI and custom actions

OpenAPI 3.1, no security scheme. The same document is at https://www.arcreport.ai/openapi.json and https://www.arcreport.ai/.well-known/openapi.json. It describes store lookup (GET /api/store), the live check (GET /api/agent-checkout), ask (GET /api/ask), the dataset manifest (GET /data/latest/manifest.json), and POST /api/outcomes, plus the other public reads.

ChatGPT actions

In a custom GPT, open Actions → Create new action → Import from URL, and paste https://www.arcreport.ai/openapi.json. Set authentication to None. GPT actions and ChatGPT apps are separate; OpenAI's help says a GPT uses one of them at a time.

Other builders

Paste either OpenAPI URL into an agent builder that imports an OpenAPI document. The old plugin manifest at https://www.arcreport.ai/.well-known/ai-plugin.json points at the same document and sets auth.type to none. It does not ask for a user account. Current ChatGPT setup is the actions import or the MCP app above.

A2A

The Agent Card follows the Agent2Agent 1.0 fields: name, description, supportedInterfaces, capabilities, skills. The current well-known path is https://www.arcreport.ai/.well-known/agent-card.json. https://www.arcreport.ai/.well-known/agent.json returns the same JSON.

Name: ARC. Description: The free, neutral record of which online stores AI agents can actually buy from. Skills: check_store, ask, report_checkout. The interfaces link https://www.arcreport.ai/api/mcp and https://www.arcreport.ai/openapi.json.

ARC does not run an A2A task endpoint. The card does not advertise JSONRPC, GRPC, or HTTP+JSON bindings. A client that only sends SendMessage will not reach ARC through those bindings. Call the skills with MCP or the HTTP API named on the card.

curl -s https://www.arcreport.ai/.well-known/agent-card.json

Plain HTTP

JSON over HTTPS. CORS is open on these reads. No key.

curl "https://www.arcreport.ai/api/store?host=allbirds.com"

curl "https://www.arcreport.ai/api/agent-checkout?url=https://allbirds.com"

curl "https://www.arcreport.ai/api/ask?q=running+shoes"

curl "https://www.arcreport.ai/data/latest/manifest.json"

curl -X POST https://www.arcreport.ai/api/outcomes \
  -H 'content-type: application/json' \
  -d '{"store_domain":"allbirds.com","agent_family":"browser","agent_name":"example-agent","timestamp":"2026-10-12T17:00:00.000Z","furthest_step":"checkout","blocker":"captcha"}'

Outcome fields and the privacy rules are on the outcome report spec. The live check is also a form at /agent-checkout.

llms.txt

https://www.arcreport.ai/llms.txt is a plain-text map of the public site: the dataset, the Index, the MCP servers, this page, and the discovery URLs. Agents that fetch a single text file can start there.

A separate note, which named agents publish MCP, OpenAPI, A2A, or browsing support, records what public pages said on 13 October 2026. Where a page did not say, the cell is unknown.

For agents · Developers · Signed-in scans