/v1/discover
Find candidate services by capability, payment rail, price ceiling, latency target, policy flags, and evidence requirement.
Agent-readable documentation draft
Docs
This public draft gives agents, LLMs, and developers the stable URLs and integration shape for pre-spend service selection, governance preflight, routing, and proof/audit readback.
Quickstart
The MCP server exposes service discovery, routing, service cards, outcome submission, and audit tools. REST also exposes governance preflight before irreversible service spend.
{
"mcpServers": {
"stackbroker": {
"url": "https://mcp.stackbroker.xyz",
"transport": "streamable-http"
}
}
}HTTP API
Fast lookup, governance preflight, routing, audit, proof readback, and telemetry ingest. No long compute. No custody of buyer funds in the MVP.
/v1/preflight
Return go, needs-confirmation, or blocked with policy checks, proof preview, net-cost delta, and context-shift flags.
/v1/route
Return seller candidates, route rationale, fallback options, policy result, and audit token; proof records persist as a side effect.
curl -X POST "https://api.stackbroker.xyz/v1/preflight" -H "content-type: application/json" -d '{"task":"fetch a page","capability":"web_retrieval","policy":{"max_price_usdc":0.05},"governance":{"require_confirmation_above_usdc":0.10}}'Service card schema
Stackbroker routes from structured service cards rather than vendor copy or public star ratings.
capabilities and endpoint URL
Used for route comparison, policy checks, procurement audit, and crawler-readable service descriptions.
auth owner and payment owner
Used for route comparison, policy checks, procurement audit, and crawler-readable service descriptions.
payment rail, price unit, quote binding, and receipt issuer
Used for route comparison, policy checks, procurement audit, and crawler-readable service descriptions.
policy flags, idempotency, replay safety, and refund path
Used for route comparison, policy checks, procurement audit, and crawler-readable service descriptions.
p50/p95 latency, schema conformance, success evidence, and last probe time
Used for route comparison, policy checks, procurement audit, and crawler-readable service descriptions.
Agent Runtime Compatibility
Stackbroker is designed as a neutral pre-spend layer for multiple agent runtimes, not a single-model marketplace. These are compatibility targets until the API and MCP endpoints return ready status.
Claude / Anthropic
MCP-first operator workflows
Claude Desktop, Claude Code-style agents, and Anthropic tool-use flows should be able to inspect Stackbroker service cards before invoking paid tools.
OpenAI / ChatGPT / Codex agents
API and tool-routing posture
OpenAI agents should be able to read the public descriptors, call discovery/routing endpoints, and retain audit rationale for paid API or MCP calls.
Google / Gemini agents
Procurement and policy metadata
Gemini-style agents need clear price, auth-owner, policy, and receipt metadata before using external data, workflow, or commerce services.
Perplexity-style research agents
Source and evidence routing
Research agents need source quality, provenance, duplication risk, freshness, and cost signals before paying for search, crawl, or data services.
OpenClaw / Hermes / internal agents
Operator-controlled spend guardrails
Internal agent fleets need routing records, budget policy checks, receipts, and reviewable audit trails before they receive real delegated budgets.
Agent Harnesses
Stackbroker should sit outside the harness as a neutral spend, routing, and audit boundary. The first compatibility watchlist focuses on five common production paths, with room for adjacent runtimes.
OpenClaw / Hermes
Self-hosted operator harnesses
Persistent personal and business agents with memory, tools, channels, scheduled work, and broad local authority need pre-spend policy, receipts, and audit trails.
LangGraph / LangChain
Stateful workflow and agent runtime
Long-running graph workflows need service-card metadata, fallback routing, and outcome telemetry at every paid external call boundary.
Microsoft Agent Framework / AutoGen
Enterprise multi-agent orchestration
Enterprise agents and graph workflows need type-safe policy gates, MCP-aware tool routing, stateful audit records, and human-in-the-loop review points.
CrewAI
Role-based multi-agent crews
Crews that delegate across roles, tools, MCP servers, apps, skills, and knowledge sources need shared procurement rules before any agent spends.
OpenAI Agents SDK
Provider-native agent handoffs and tools
OpenAI-first agents need a neutral service selection layer when tool calls leave the model provider and touch paid APIs, data, or MCP services.
Also relevant: Google ADK, Mastra, LlamaIndex Workflows, Vercel AI SDK, Pydantic AI, Claude Agent SDK / Claude Code, Deep Agents SDK, Temporal, and Inngest.
MCP tools draft
Stackbroker will start with a small tool surface designed for agent procurement.
| discover_service | Return ranked candidate services for a requested capability. |
|---|---|
| route_request | Return seller candidates, route rationale, policy fit, payment metadata, receipt expectations, and fallback options. |
| get_service | Fetch a normalized service card, verification history, attestation tier, and payment posture. |
| submit_verification_result | Record a probe result, real task outcome, latency sample, schema conformance result, or failure signal. |
| audit_route | Return route rationale, receipt references, outcome telemetry, and log references for a prior decision. |
x402 posture
Stackbroker is being designed around x402-style agent payments while staying routing-only at launch.
Sellers expose payable endpoints. Stackbroker returns routing metadata, price claims, quote-binding hints, verification evidence, receipt expectations, and payment instructions. Buyers pay sellers directly unless and until a later compliant transaction architecture is approved.
Trust pre-flight (signed manifests)
Point-in-time trust evidence on every scanned card — never a certification. Full scope at /trust.
Scans cover what a service declares: static description/schema analysis (injection-pattern and credential-harvesting content), schema-drift detection against daily snapshots, and endpoint reputation (TLS/HTTPS/DNS). Explicitly out of scope: runtime traffic inspection and novel attacks. Findings map through a versioned rubric — HIGH severity requires human review before any public effect — and results ship as Ed25519-signed manifests at
GET /v1/services/{id}/trust, with last_scanned_at / scan_status / manifest URLs on every discover and route candidate. Manifests expire and display as stale, never hidden. Scanning is free to providers, always: neutrality policy.Outcome telemetry
The useful review primitive is a structured record from a real routed or probed task.
| Task result | downstream success, retry count, parse repair, schema conformance, and uniqueness versus free sources |
|---|---|
| Cost result | quoted price, actual price, duplicate-charge risk, refund path, and receipt issuer |
| Quality result | latency, provenance quality, policy flags, operator note, and recommendation |
Status
Public uptime and route-quality reporting will live at https://status.stackbroker.xyz. For now, this page is the canonical public status placeholder.
| Apex site | Live |
|---|---|
| Docs | Public draft |
| Directory | Seed catalog draft |
| API/MCP | Live proof-slice with discovery, preflight, routing, audit, outcomes, probes, and MCP tools |
| Evidence loop | Started with one real paid x402 Orbis call; broad benchmark data is not claimed yet |
Reserved surfaces
docs, directory, api, mcp, app, admin, status, and stage hostnames are mapped or reserved. Admin is intentionally not public, and stage is protected for review.