SBStackbroker
The neutral pre-flight layer for agent spend
GO.
HOLD.
NO-GO.

Every agent purchase gets a pre-flight check. Stackbroker is the layer an agent calls before it spends — policy checks, a reasoned route, and a sha256-stamped proof for every decision. Routing-only. No custody.

Operators: get clearanceAgents: one call firstProviders: claim your card
PRE-FLIGHT · POST /v1/preflightPOLL 1/3
task: Fetch and render 40 supplier pages
capability: web_retrieval
policy: max_price_usdc: 0.01 · blocked_policy_flags: [pii_processing]
  • 01 candidate_available — 3 listed servicesPASS
  • 02 policy_flags_clear — no blocked flagsPASS
  • 03 spend_within_approval — $0.004 ≤ $0.01PASS
  • 04 attestation_tier_met — seed acceptedPASS
polling policy checks…
Pre-flight checklist — how a spend decision clears

Four items. Every paid call.

  1. 01

    The call

    Before paying any API, MCP tool, data feed, or x402 endpoint, the agent makes one call to Stackbroker with its task, the capability it needs, and the spend policy it operates under.

    POST /v1/preflight · POST /v1/route · MCP route_request

  2. 02

    The checks

    The request is polled against real policy: budget caps, allowed payment rails, blocked capability flags, attestation floors, confirmation thresholds, spend anomalies, and context shifts.

    candidate_available · spend_confirmation_required · approved_spend_exceeded · governance_service_blocked · spend_anomaly_threshold · service_context_shift

  3. 03

    The decision

    A deterministic 0–100 score ranks every candidate — task success, schema conformance, cost per successful task, p95 latency, provenance — and the board returns GO, HOLD for confirmation, or NO-GO. With reasons, a second choice, and fallbacks. Never a black box.

    decision: go | needs_confirmation | blocked · why_selected[] · risk_flags[]

  4. 04

    The proof

    Every decision is persisted with an audit ID and a content-addressed proof hash. Finance and governance can read the trail back at any time — what was asked, what was checked, what was chosen, and why.

    audit_id · proof_hash: sha256:… · GET /v1/audits/{id} · GET /v1/governance/events

Three doors

Who calls the tower.

For humans deploying agent forcesOperators

Your agents are getting budgets. Get pre-flight confidence: hard policy boundaries, spend caps, and an audit trail you can hand to finance.

Get clearance →For agents and the developers wiring themAgents

Don't waste a paid call choosing poorly. One call before you spend returns the best route, the price, the policy verdict, and the receipt.

GET /llms.txt →For services agents payProviders

Agents can't choose you from a marketing page. Claim the service card that makes you routable — capability, price, rails, policy, evidence.

Claim your card →

Service cards, not star ratings.

Every service is one normalized data plate — readable by humans, exportable to agents, usable for a policy check before money moves.

Firecrawltier: seed
service_idsvc_firecrawl
providerfirecrawl.dev
capabilitiesweb_retrieval · scraping
payment_railsapi_key
price_unitper_call
policy_flagsexternal_network
p95_latency_ms— (null)
success_rate— (null)
Evidence fields stay null until probes or paid telemetry produce them. We do not invent benchmarks.

The Megaboard.

One neutral board for the agentic web — agent-callable services synthesized from multiple public catalogs into a single evidence-labeled ranking. The first of several Stackbroker benchmarks.

Top services · Stackbroker Megaboard (megaboard_v0)
#ServiceScoreEvidence
1CoinMarketCapcoinmarketcap.com52observed
2Tavily Search - advanced modex402.tavily.com52observed
3Exa /search endpointapi.exa.ai47observed
4QuickNodex402.quicknode.com47observed
5Search onchain DEX pools and token, by contract address, name, or token symbol.pro-api.coingecko.com47observed
See the full MegaboardEmbed it

One real paid call. No invented benchmarks.

The governed spend loop is live with one real paid x402 call. Routing-only: buyers pay sellers directly, and Stackbroker holds nothing.

Launch contract
Endpointshttps://api.stackbroker.xyz/v1/discover · /v1/preflight · /v1/route · /v1/audits · /v1/outcomes — REST and MCP
Decision statesgo · needs_confirmation · blocked — with itemized policy checks and a proof hash on every decision
Payment posturex402-aware metadata and buyer-to-seller payment instructions. No custody in the launch architecture.
Evidence postureService-card evidence fields stay null until probes or outcome telemetry exist. Honest by contract, not by promise.
llms.txtopenapi.json/.well-knownMCP endpoint

Verifiable neutrality.

Paid by the demand side — agents subscribe; providers can't pay us, so they can't tilt us. Neutrality you can verify: a published methodology, signed audit anchors, and a money model with only one side.

Check the math yourself
MethodologyEvery scoring factor, every weight, and what is never a factor — the same versioned config the router runs on, drift-tested in CI.
Audit anchorsEvery decision hash-chained; daily Ed25519-signed anchors published with a verification walkthrough.
Neutrality policyProviders never pay — no listing, verification, or placement fees. Subscriptions buy access, never outcomes.