What changed
Two tools shipped the same quarter: swaggertomcp/MCP Gateway turns any OpenAPI spec or Postgres schema into an agent-callable MCP server with no custom code, and Cloudflare's Monetization Gateway attaches per-request x402 stablecoin billing to any resource behind its edge. A working example (goldbean-api, 47 endpoints, x402 micropayments) is already live.
Why now
The wrapper (generate the server) and the payment rail (meter per call) both became available and proven this quarter, and OpenAI's push toward long-running autonomous agents is expanding the population of machine buyers that consume structured inputs. FACT: all three source artifacts are dated to this quarter per the input. HYPOTHESIS: that autonomous agents actually pay for third-party data tools at scale is unproven.
Converging signals
swaggertomcp (API/DB→MCP server) + Cloudflare Monetization Gateway (x402 per-request settlement) + a flood of long-running agents needing trustworthy structured inputs, with goldbean-api as an existence proof of the pay-per-call MCP pattern.
Customer pain
Vertical-agent builders and enterprises exposing internal data need trustworthy, normalized domain data as agent-callable inputs; assembling and maintaining clean datasets (e.g. fragmented federal/state grant and reimbursement rules) is painful. INFERENCE — no demand_evidence (complaints, job posts, or paying customers) was supplied, so this pain is asserted from the founder's own dataset-building experience, not proven.
Who pays
Builders/operators of vertical AI agents and enterprises wiring internal systems to agents, paying per query in stablecoin via x402. The BENEFICIARY (agent that gets clean data) and BUYER (dev who runs the agent) are the same party here.
Solved today
Agents either scrape/parse raw sources themselves, hand-curate a dataset, or pay for a conventional API-key/subscription data product. No widespread per-call MCP data marketplace exists yet.
Why current solutions are bad
DIY scraping is brittle and expensive per agent; subscriptions don't fit ephemeral per-task agent usage. But 'bad' is mild — the free/DIY path is often adequate, which is the core risk.
Proposed product
A small catalogue of paid MCP data tools over datasets the founder can uniquely curate: structured, normalized, deadline-tagged federal+state grant/claimable-money/FEMA-PA reimbursement rules, queryable by agents and billed per call via x402. Lead with ONE endpoint over the hardest-to-assemble dataset.
MVP version
Curate and normalize one dataset into Postgres; run swaggertomcp to generate the MCP server; front it with Cloudflare Monetization Gateway/x402; publish to MCP registries/directories; instrument real paid calls.
30-day build
Pick the single dataset with the clearest 'painful to assemble' moat (leverage existing grants/claimable-money curation work). Normalize it, ship the MCP endpoint behind x402, list it in MCP directories, and run the KILL TEST: seek a handful of independent paid calls.
60-day build
If paid calls materialize, add 2-3 adjacent endpoints and a conventional API-key/subscription fallback for buyers who won't touch stablecoin; publish usage examples and court a few vertical-agent builders directly.
90-day revenue plan
Convert early callers into recurring committed usage (prepaid credits or subscription tier over the same data); pursue a white-label/data-licensing deal with one agent platform as the durable revenue path rather than relying on spot micropayments.
Distribution path
MCP tool registries/directories, Show HN, agent-builder communities, and direct outreach to vertical-agent operators. Distribution to agent-side buyers is the single biggest unknown.
Pricing hypothesis
Per-call micropayment (e.g. $0.002-$0.02/query) via x402 for spot usage; prepaid credit packs or a $99-$499/mo subscription for steady consumers; negotiated data-licensing for platforms.
Technical difficulty
Low on plumbing (both server generation and billing are off-the-shelf); the real work and the moat are dataset curation, normalization, and freshness — which is ongoing labor, not a one-time build.
Legal / regulatory risk
Low-moderate: curating public data is fine, but verify redistribution terms of any source; x402 stablecoin settlement adds light money-transmission/tax bookkeeping considerations. No licensure required.
Platform dependency
HIGH and honest: this bets on Cloudflare's x402 rail, the MCP ecosystem, and agents willing to pay machine-to-machine. If x402/agent-payments adoption stalls, the per-call model has no buyers — the conventional API fallback is the hedge.
Founder fit
Partial. The DATASET (structured public-money/grants/reimbursement rules) sits squarely in the founder's proven public-money wheelhouse and reuses curation he's already doing. But the BUSINESS SHAPE — selling speculative machine-payment access to agents — is NOT his forced-filer/government-portal sweet spot and depends on an unproven buyer, so founder_fit is moderate, not maximal.
Breakout potential
Real if agent-payments take off: a catalogue of authoritative, agent-callable public-money data tools could become default infrastructure. But it's a bet on a nascent rail, not a compelled-filer certainty.
Final recommendation
WEAK PURSUE / cheap probe. Don't build a catalogue. Run the input's own kill test: ship ONE paid MCP endpoint over the founder's hardest-to-assemble public-money dataset (reusing existing curation) in ~2 weeks and watch for real independent paid calls. Proceed only if demand shows; otherwise the curated dataset still has value sold the founder's proven way (per-filing/subscription to human operators).
Next action
Select the single most defensible existing dataset, generate its MCP server via swaggertomcp, put it behind Cloudflare x402, list it, and measure real paid calls over 2-3 weeks.