Site intelligence for agents, over MCP

Pick your client and run the line. The client opens a browser so you can sign in, or create an account, and approve access. There is no key to paste, and the free plan needs no card.

Any agentClaude CodeCodexGemini CLIClaude and ChatGPT apps

npx add-mcp https://askatria.ai/mcpCopy

Detects the agents on your machine, such as Claude Code, Cursor, VS Code, Codex and Gemini CLI, and adds Atria to each. Your client opens a browser to sign in and approve; the free plan needs no card.

For any other MCP client, add the server to its configuration file:

{
  "mcpServers": {
    "atria": {
      "type": "http",
      "url": "https://askatria.ai/mcp"
    }
  }
}

Then ask your agent something about a place, for example: “Who owns 4405 Michigan Ave Rd NE in Cleveland, Tennessee, and how much of it is in the floodplain?”

OAuth is the default for interactive clients. Atria supports dynamic client registration, so a client that speaks MCP authorization registers itself, opens the sign-in page and keeps its own tokens. The metadata is at /.well-known/oauth-authorization-server and /.well-known/oauth-protected-resource.

API keys are for servers, scripts and clients without OAuth. Create one under Scout, then Agents in your dashboard and send it as a bearer token. Keys start with atria_sk_ and are shown once.

{
  "mcpServers": {
    "atria": {
      "type": "http",
      "url": "https://askatria.ai/mcp",
      "headers": { "Authorization": "Bearer atria_sk_..." }
    }
  }
}

Keys carry scopes: lookup, screen, search (which also covers atria_result), coverage and crawl. A call outside the key’s scopes is refused and costs nothing.

The server speaks MCP over streamable HTTP and is stateless: every request is a JSON-RPC message POSTed to the endpoint. MCP clients handle this for you; without one, list the tools and call them directly.

curl -s https://askatria.ai/mcp \
  -H "Authorization: Bearer $ATRIA_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

curl -s https://askatria.ai/mcp \
  -H "Authorization: Bearer $ATRIA_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "atria_lookup",
      "arguments": { "address": "4405 Michigan Ave Rd NE, Cleveland, TN" }
    }
  }'

GET /mcp/info returns the server’s name, version and tool list without authentication.

Every value is wrapped in an envelope so an agent can cite it and can tell a real gap from a source that did not answer:

• ok: the value, with source (name and URL) and fetched_at.
• absent: the source answered and has nothing at this location, with a reason.
• failed: the source did not answer. This says nothing about the site; retry later or check another source.

{
  "status": "ok",
  "value": { "zones": [{ "zone": "A", "acres": 4.29, "percent": 7.2 }], "sfhaPercent": 7.2 },
  "source": { "name": "floodZone", "url": "https://services.arcgis.com/.../FeatureServer/0" },
  "fetched_at": "2026-10-01T00:00:00.000Z"
}

The server instructs agents never to fill an absent or failed value in from general knowledge.

Pass any of these in atria_screen’s modules. National modules cost 10 credits and county modules, which read the county’s own GIS, cost 25.

Calls are priced in credits; 1 credit is $0.001. Credits belong to your organization and reset on the first of each month (UTC). A call that would exceed what is left is refused and costs nothing.

Manage your plan in the dashboard.

• 401 with a WWW-Authenticate header: no valid token or key. OAuth clients use the header to start sign-in.
• Insufficient credits: the call would exceed your organization’s remaining credits. Nothing is charged; credits reset on the first of the month, or raise your plan.
• Rate limits apply per key per minute. A limited call is refused and costs nothing.
• Tool errors come back as MCP tool results with isError set and a message the agent can read.