Market intelligence your AI agent can call: 7 MCP tools, priced per call

Most market-data products assume a human will sign up, generate an API key, read the docs, and wire up a client. AI agents don’t work that way. NataPulse now serves its curated market intelligence — published events, emerging narratives, citable research reports — through a read-only MCP server at mcp.nata.my, with seven tools priced individually per call in USDC, from $0.005 to $0.05. Discovery is free. There is no account to create and no API key to provision: the design goal is that an agent connects, lists the tools, and pays only for the answers it actually retrieves.

One endpoint, stateless by design

The server speaks the Model Context Protocol over Streamable HTTP: a single POST https://mcp.nata.my/mcp endpoint carrying JSON-RPC 2.0, with no session state. Every request is self-contained, which means standard MCP clients — Claude, Cursor, mcp-remote — connect without extra configuration, and the surface behaves predictably under retries and parallel calls. A public health check lives at /healthz, no auth required.

Statelessness is not an implementation detail; it is what makes the per-call pricing model coherent. An agent doesn’t hold a subscription or a session — it makes a request, and that request is the entire unit of commerce.

The seven tools

The surface is deliberately small. Seven tools cover the three things NataPulse publishes:

list_events ($0.005) — curated, clustered market events, published only.
get_event ($0.005) — full detail of one event: body, public scores, citable sources with direct links, cluster context and related public reports.
list_narratives ($0.02) — emerging market narratives detected across sources, with public score, trend, momentum and a 12-point sparkline.
get_narrative ($0.03) — one narrative in depth, plus the event clusters that feed it.
list_reports ($0.005) — generated intelligence reports (catch-up, event, ticker, theme and more).
get_report ($0.05) — the full report: markdown body, references and citations, quality breakdown of published versions.
search ($0.005) — keyword search across published events, clusters and reports.

The price bands track derived value, not payload size. Lookups and lists — the calls an agent makes to find an identifier — sit at the $0.005 floor. A full narrative costs $0.03; a complete cited report, the densest artifact NataPulse produces, sits at the $0.05 ceiling. list_reports is priced as a lookup rather than as content, because its job is to find a report ID; the value lives in get_report.

Paying per call: x402, rolling out

Payment uses the x402 protocol — the HTTP 402 status code, revived as a machine-payments standard. The flow is three steps: call a priced tool with no payment and receive a 402 with a PAYMENT-REQUIRED challenge; sign a payment authorization and retry the same request with a PAYMENT-SIGNATURE header; receive the result along with a PAYMENT-RESPONSE carrying the on-chain receipt. Settlement is in USDC on Base, non-custodial — funds move atomically from payer to merchant. Payments are final; there are no chargebacks, though a tools/call against a nonexistent tool fails locally and costs nothing.

One thing to be precise about: x402 billing is built but not yet switched on. Today, discovery and the health check are public, and tool access is granted on request. The published price table is the contract agents will find at activation. Per-wallet defaults at launch: 2,000 calls per day, $10 per day, 30 requests per minute — higher limits on request.

Everything before payment is free by rule, not by grace: initialize, tools/list, ping and notifications never cost anything, because an agent must be able to inspect a surface before deciding to pay for it.

Machine-readable, agent-first

Alongside the human-readable /for-agents page, NataPulse publishes two machine-readable twins: /llms.txt, a structured summary of the surface, tools and prices that a language model can ingest directly, and /AGENTS.md, with connection instructions, the payment flow and worked examples. All three are generated from the same source of truth, so the prices an agent reads are the prices the server enforces.

This is the practical case for agent-native distribution. The traditional funnel — account, email verification, key management, monthly plan sizing — is friction designed for humans and organizations. An agent doing research on behalf of a user may need exactly four calls: one search, one event, one narrative, one report. Under per-call pricing that unit of work has a knowable, sub-dime cost, and the wallet is the identity.

Read-only by construction

The MCP server can only read published, public data. It has no write path, no access to unpublished or internal content, and its outputs pass through a double layer of sanitization before leaving the server — the same publication boundary that governs what appears on natapulse.com. Agents get the evidence layer: events with provenance, narratives with explainable public scores, reports with citations. What they do not get — from this surface or any other — is trade instructions. NataPulse produces research evidence, not advice, and that framing ships in the payloads themselves.