MCP Server
Give an agent
your documents.
Fourteen tools built around what someone actually asks an underwriting agent to do, not 210 generated from endpoints. Read-only by default.
Connect
{
"mcpServers": {
"moneyline": {
"command": "npx",
"args": ["-y", "@moneyline/mcp"],
"env": { "MONEYLINE_API_KEY": "sk_live_..." }
}
}
}In practice
{
"submissionId": "sub_a1b2c3",
"income": { "monthlyMinor": 762000, "confidence": 0.94 },
"cashflow": { "avgDailyBalanceMinor": 1841233, "nsfCount": 0 },
"positions": [{ "funder": "Example Capital", "dailyMinor": 41500 }],
"flags": [],
"documents": ["doc_9f8e7d"]
}One tool per intent
analyze_submission runs parse, enrich, and score internally. Every step the agent does not orchestrate is a step it cannot get wrong.
Summaries, not payloads
Arithmetic is done in the tool and handed over as a number. A 40-page statement never lands in the context window.
Read-only unless you say otherwise
The four tools that write are gated behind an explicit key scope, so a token issued for analysis cannot approve anything.
Document text is marked untrusted
A submitted PDF can contain text addressed to the agent reading it. Parsed content comes back in a delimited block flagged as untrusted input.
Start with
one document.
Install, authenticate, and parse your first file. No credit card, no sales call.