MCP quickstart
Three steps take an agent from no account to a first answer from Arche's MCP
server at https://api.arche.fi/mcp: get a key, connect a client, call a tool.
Get a sandbox key
An agent can open a Free account for its owner and receive a sandbox key in the same response, with no browser step:
curl -X POST "https://api.arche.fi/v1/agent/signup" \
-H "Content-Type: application/json" \
-d '{"email": "owner@yourcompany.com", "agent_name": "research-assistant"}'
The key is returned once as data.api_key, so store it at once. A sandbox key
reads the same data as a production key. The owner is emailed and can take the
account over later; see Agent signup. Export the key for the
steps below:
export ARCHE_API_KEY="ak_test_…"
Connect Claude Code
claude mcp add --transport http arche https://api.arche.fi/mcp \
--header "X-Api-Key: $ARCHE_API_KEY"
Then ask Claude something the tools answer, for example "What were Apple's total assets on its latest balance sheet, and which filing are they from?"
Connect Cursor
Add the server to ~/.cursor/mcp.json, or to .cursor/mcp.json in a project:
{
"mcpServers": {
"arche": {
"url": "https://api.arche.fi/mcp",
"headers": { "X-Api-Key": "YOUR_API_KEY" }
}
}
}
Keep the key out of configuration you commit.
Call it over HTTP
Any HTTP client can call the server with JSON-RPC 2.0. Resolve a ticker:
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/call", "params": {"name": "companies_resolve", "arguments": {"ticker": "AAPL"}}}'
The result's content[0].text holds the tool's JSON output, here Apple's CIK,
0000320193. Pass that CIK to financials_get_as_of_snapshot to read its
statements as they were known on a date, with a replay_token that
replay_verify checks later.
Next
- MCP: every tool, its arguments and plan, errors and limits.
- Python SDK: the same data from Python.
- Verifiable agent answers: have an agent cite and verify what it read.