October 7, 2026 · ScriptMasterLabs · the x402 / MCP / AI-agent pedia

Fund Momentum's pay-per-call MCP server: every data lookup is a payment decision

On October 6, 2026, Vienna's Fund Momentum opened its VC and LP database to AI agents through a remote MCP server with machine payments in USDC on Tempo: keyless trial, one-call self-registration, pay per call from €0.01, no account required. The announcement is the cleanest live specimen yet of the pattern this site tracks — every tool call is a tiny machine payment, and the vendor just deleted the confirmation step. The direct answer: the provenance block gives the agent something to evaluate, but nothing on the buy path scores whether this call should fire. That is the missing layer.

The receipts, dated

The missing layer: nobody scores the buy

Fund Momentum did the hard parts right. The 402 challenge is self-describing — it tells the agent exactly what to pay, how much, and on which rail. The receipt is a claim atom: settlement evidence, not a promise. The no-double-pay rule and the wrong-slug-rejected-before-challenge rule are real settlement discipline. And the provenance block is genuinely novel as data infrastructure: source, last-verified date, confidence score on every record, designed for the agent as the evaluator.

But follow the money, not the schema. 25 free credits makes the first paid call feel like fetching, not buying — then the 26th call spends real USDC. Every tool is declared read-only and idempotent, so the vendor's own confirm band is set to zero: no confirmation step between the question and the payment. At €0.01 a lookup, a bored agent can run a 4,000-call sweep of match_startup at €0.25 a pop — that's €1,000 of machine spend with zero scored judgment between intent and settlement.

The provenance block is half of the answer the gate needs. A confidence score on the data tells the agent whether the record is trustworthy; a scored gate tells it whether this call deserves to fire. Source, freshness, confidence-in-data × instruction, budget, provenance-of-the-question: that fusion is the decision. Fund Momentum hands the agent the data half. Nobody hands it the gate half.

The live gate test, Oct 7, 2026

Two instructions scored against scriptmasterlabs.com/api/harness/decide this morning (local-heuristic-v1, bands ≥0.80 auto / 0.50–0.79 confirm / <0.50 escalate):

• "Should the agent pay €0.01 via MPP on Tempo for one search_funds call to a VC-data MCP server for a founder raising a seed round, the tool is listed as read-only and the record carries a provenance block with a fresh verification date?" → 0.35, escalate/block+log
• "Should the agent auto-fire 4,000 match_startup MCP tool calls at €0.25 each (€1,000 total, no per-call review, no user approval) to a VC-data server for a broad unfocused sweep after its 25 free MPP credits ran out?" → 0.35, escalate/block+log

The blunt heuristic escalates both identically — it cannot discriminate a €0.01 listed-price lookup from a €1,000 credit-drain sweep. Fail-closed blocks both. That is the eighteenth consecutive run with this finding, and the tension stands: a calibrated decider would green-light the €0.01 provenanced call at ≥0.80 and block the €1,000 sweep below 0.50. "Block everything" is a seatbelt, not judgment.

Do it yourself: 5 steps

1. Consume the provenance block. Don't rank records by relevance alone — weight source × freshness × confidence before the question ever reaches a buy decision.
2. Score the call, not the server. Trust in the tool (read-only, idempotent) is not trust in this invocation. Band it: listed-price ≤€0.01 routine buy → auto; €0.10–0.25 per call → confirm; unbounded sweeps → escalate.
3. Put the per-call price in the envelope. Per-task budgets: cap a research task at N calls × price. Fund Momentum's get_changes incremental-sync tool is the disciplined pattern — pull the diff, don't re-poll the world.
4. Log like the FTC is watching. Keep the receipt from every 402 retry: challenge id, amount, rail, band, decision, timestamp. Machine payments need auditable records of instruction/authority/intent/outcome.
5. Never outsource judgment to "no confirmation needed." Read-only and idempotent describe the side effects, not the spend. The confirmation you skip is the payment you authorized.

Claim Receipts

Built for this lane

• Decision-Gated Payments — the pattern: confidence bands ≥0.80 auto-pay, 0.50–0.79 confirm, <0.50 escalate, wired to x402.
• AI Agent Spending Limits — enforce the cap at the payment layer, not the prompt; per-call price envelopes for pay-per-call MCP servers.
• x402 Agent Spending Guard — guard-vs-gate: the ceiling answers "can we afford it," the gate answers "should it happen at all."
• Monetize an MCP Server — the seller side: x402 v2 headers, real prices, live payment proof on a production MCP server.
• x402 Payment Protocol — the 402 Payment Required rail: how the challenge flow actually works, with live receipts.
• Provider Trust — verifying the provider behind the data: first-party provenance you can audit.

FAQ

What is Fund Momentum's MCP server? A remote MCP server (announced Oct 6, 2026) exposing Fund Momentum's VC/LP fundraising database to AI agents: 1,100+ VC funds, ~300 fund-closing research reports, 650+ LPs. Agents connect at fundmomentum.vc/_api/mcp with Claude, ChatGPT, Cursor, or any MCP client.

How does an agent pay for Fund Momentum data? Keyless trial, one-call self-registration (25 free credits), then pay per call in USDC on the Tempo network via the Machine Payments Protocol: the paid tool returns HTTP 402 with a payment challenge, the agent pays and retries, gets data plus a receipt.

How much does pay-per-call cost? €0.01 per call for search_funds/get_fund/get_changes, €0.10 for get_fund_signals/get_gp_profile, €0.25 for match_startup. Free tier: 100 calls/month. LP records (search_lps) are not sold per call — they need the LP Radar subscription.

What is the provenance block? Every record carries its source, the date it was last verified, and a confidence score — metadata designed so an AI agent can evaluate trust. It feeds a decision gate's data-confidence input, not the buy decision itself.

Is the no-confirmation design safe? Read-only and idempotent means the call can't change anything server-side — it doesn't mean the payment was worth making. With no confirmation step, every call fires real USDC. Score each call's worth separately from the tool's safety.