I Built a Verified LEI + FIGI Lookup Agent for $5/month — Callable by Any AI Agent via x402

Here is what I built: a financial data normalization agent that any autonomous AI agent can discover and pay for in USDC on Base — with zero signup, zero API keys, and zero human involvement.

It went from idea to live production in one afternoon, entirely from my iPhone.

The Problem It Solves

If you are building a trading pipeline, compliance agent, or portfolio tool, you constantly deal with messy financial data:

Dates in a dozen formats
Currency written as “USD”, “usd”, “$”, “dollars”
Counterparties identified by nickname, abbreviation, or internal code
No ISIN, no LEI, no FIGI — just a ticker that may or may not be unique

Before your agent can act on this data, something has to clean it. That something is expensive to build yourself and boring to maintain. So I built it once and made it callable by any agent.

What It Does

The agent exposes four tools over a standard HTTP API:

normalize_financial_record — Takes any raw trade, transaction, or position record in any format and returns structured JSON with standardized dates (ISO 8601), currency codes (ISO 4217), inferred asset class, and data quality flags.

validate_regulatory_fields — Checks a record against MiFID II, Dodd-Frank, EMIR, or SEC Rule 605 and returns a list of missing or invalid fields.

enrich_counterparty — Given a company name, returns a verified LEI code, jurisdiction, and entity status from the GLEIF Global LEI Index — the same database used by regulators and central banks worldwide. Zero AI inference. Real verified data.

lookup_instrument — Given an ISIN, CUSIP, SEDOL, or ticker, returns a verified FIGI code, exchange, and security type from OpenFIGI. Auto-detects the identifier type.

The Part That Makes It Different

The counterparty enrichment and instrument lookup tools do not call an AI model. They call real financial databases directly:

enrich_counterparty("Goldman Sachs International")

→ {
    "lei": "549300R83Q2VTICWNV15",
    "legal_name": "GOLDMAN SACHS INTERNATIONAL",
    "jurisdiction": "GB",
    "status": "ACTIVE",
    "verified": true,
    "source": "GLEIF"
  }

That LEI code is pulled live from GLEIF at query time. It is the same code regulators use for MiFID II reporting. No LLM could hallucinate its way to that with confidence.

This matters because it means those two tools cost essentially nothing to run — no AI tokens, no model inference, just a free API call. The margin on counterparty enrichment calls is close to 100%.

The Stack

FastAPI on Python — the server
Claude Haiku — cheapest capable model, ~$0.00025 per call
GLEIF API — free, no key required, returns verified LEI data
OpenFIGI API — free, returns verified FIGI data
Railway — deployment, ~$5/month
Stripe — metered billing for API key customers
x402 — USDC micropayments on Base for autonomous agents

Total infrastructure cost at low volume: under $10/month.

Two Payment Rails

This is the part I find most interesting architecturally.

The agent accepts payment two ways:

Stripe metered billing is for developers who want to integrate the agent into their pipeline the traditional way — sign up, get an API key, pay monthly. Standard SaaS pattern.

x402 is for autonomous agents. Any agent with a funded wallet on Base can call the endpoint right now, with no signup required. The server responds with HTTP 402 + payment instructions, the agent pays in USDC, and the data comes back. No humans involved at any step.

GET https://your-agent.railway.app/.well-known/x402

→ {
    "version": 2,
    "resources": {
      "/invoke": {
        "price": "$0.002",
        "network": "eip155:8453",
        "token": "USDC"
      }
    }
  }

The price per call is $0.002 regardless of which rail the caller uses. Consistent pricing, two very different buyer journeys.

Discovery Endpoints

The agent is machine-readable at multiple standard endpoints so orchestrators and discovery services can find it automatically:

GET /tools                          — MCP tool listing
GET /.well-known/agent.json         — Google A2A manifest
GET /.well-known/x402               — x402 payment manifest
GET /.well-known/mcp.json           — MCP discovery (upcoming standard)
GET /health                         — Status + active payment rails

Any agent that knows to look at /.well-known/agent.json or /.well-known/x402 can discover this agent, understand what it does, and pay for it — without any human writing documentation or sending an email.

The Economics

Here is the per-call cost breakdown at the current price of $0.002:

Tool	Cost to run	Gross margin
normalize_financial_record	~$0.00025 (Haiku)	~87%
validate_regulatory_fields	~$0.00025 (Haiku)	~87%
enrich_counterparty (GLEIF hit)	~$0.00001 (API call)	~99%
lookup_instrument (OpenFIGI hit)	~$0.00001 (API call)	~99%

At 10,000 calls/month the price drops to $0.0015/call. At 10,001+ it drops to $0.001. Infrastructure cost stays flat. Margin improves at scale.

How I Built It on iPhone

The entire build happened from my iPhone. No laptop, no VS Code, no local terminal.

Code editing: Textastic (iPhone app)
Deployment: Railway connected to GitHub, auto-deploys on commit
API testing: API Tester app
Billing setup: Stripe Dashboard mobile
Monitoring: Railway logs in Safari

The main challenge was iPhone autocorrect converting straight quotes to curly quotes in Python code, which causes syntax errors. The fix was using single quotes throughout the codebase — iPhone never autocorrects single quotes.

What Is Next

The agent is live on Nevermined, listed in the MCP registries, and discoverable via x402. The next meaningful upgrade is adding OpenFIGI batch processing — the API supports up to 100 identifier lookups per call, which would make bulk trade normalization dramatically faster and cheaper.

The longer-term bet is that financial compliance pipelines become a high-volume source of A2A transactions. Every trade that goes through a reporting agent needs normalized data. That is a large number of calls, and the agents running those pipelines will eventually have wallets.