MCP

Arche runs a Model Context Protocol (MCP) server, so an AI agent can look up companies, read normalized statements and point-in-time financials, and check data quality without custom integration code. Each tool calls the same HTTP route your code would, under your API key, so an agent gets the same figures, the same provenance and the same plan coverage.

MCP is available on every plan, including Free. A Free key reaches the same 25 companies it can query over HTTP. Modeling, reconciliation and AI tools need a paid plan; the tool table lists the plan each one needs.

Connect a client

Any client that can connect to a remote MCP server over HTTP with a custom header can use Arche. With Claude Code, for example:

claude mcp add --transport http arche https://api.arche.fi/mcp \
  --header "X-Api-Key: YOUR_API_KEY"

Clients configured from JSON generally take the same three values:

{
  "mcpServers": {
    "arche": {
      "type": "http",
      "url": "https://api.arche.fi/mcp",
      "headers": { "X-Api-Key": "YOUR_API_KEY" }
    }
  }
}

Keep the key out of shared configuration files. Most clients can read it from an environment variable instead.

Call it directly

The server speaks JSON-RPC 2.0 and supports initialize, ping, tools/list and tools/call. Notifications such as notifications/initialized are acknowledged with 202 and no body.

List the tools:

curl -X POST "https://api.arche.fi/mcp" \
  -H "X-Api-Key: $ARCHE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}'

Each tool carries a name, a description and an inputSchema. Call one with its arguments:

curl -X POST "https://api.arche.fi/mcp" \
  -H "X-Api-Key: $ARCHE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "id": 2, "method": "tools/call", "params": {"name": "companies_resolve", "arguments": {"ticker": "AAPL"}}}'

The result is the tool's output as JSON text:

{
  "jsonrpc": "2.0",
  "id": 2,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "{\"as_of\": \"2026-09-26\", \"cik\": \"0000320193\", \"exchange\": null, \"request_id\": \"58c0f33bcff243b9a4cd83ebc36b89df\", \"ticker\": \"AAPL\"}"
      }
    ],
    "isError": false
  }
}

Tools

Tool names are the method names with _ in place of .. Every tool is available on every plan unless the table says otherwise; a tool above your plan returns a FEATURE_NOT_AVAILABLE result naming the plan it needs. The same list, with each tool's input schema, is served at GET /mcp/manifest.

The modeling_* and ai_* tools take company_id, a UUID, rather than a CIK. Get one from /v1/identity/resolve. The modeling tools also need a snapshot for their as_of date; see Modeling.

Errors

Two kinds of failure come back differently.

A malformed request is a JSON-RPC error, returned with HTTP 200:

A tool that runs and fails, because of your plan, a rate limit or the route it calls, returns a result with isError: true. Its text is a JSON error:

{
  "type": "FEATURE_NOT_AVAILABLE",
  "message": "This feature requires the growth plan or higher.",
  "retryable": false,
  "http_status": 403,
  "http_code": "FEATURE_NOT_AVAILABLE",
  "trace_id": null,
  "retry_after_s": null
}

Retry only when retryable is true, and wait retry_after_s seconds when it is set. Quote trace_id when you escalate.

Limits

MCP requests are metered in their own class; see Rate limits for your plan's budget, which the manifest also reports under rate_limits. A request body larger than 1 MiB is refused with 413.