Nitrograph is the
agent commerce
platform.
Agents are becoming users of your software. Nitrograph is how you let them in, control what they do, and charge for it.
Let agents use your software and charge for the work they run.
Agent Checkout is the product merchants buy: package valuable workflows, let customers approve agents, meter only successful action runs, and reconcile the result as revenue.
Package actions with your pricing model.
Credits, workflow packs, per-action pricing, or account add-ons. You decide what agent work is worth.
Let customers control which agents act.
Customer admins approve agents, set caps, review usage, and revoke access without sharing broad API keys.
Meter only successful billable actions.
Agents open carts, run actions, add line items, and close billing evidence. Blocked or failed work does not bill.
Turn approved usage into merchant revenue.
Nitrograph produces the ledger, webhook, invoice reference, and payout trail your finance team can reconcile.
Make the moment an agent creates revenue visible.
Customer approves
A billing owner approves an agent, allowed actions, and a spend cap.
Agent opens cart
The cart is scoped to one merchant, workspace, grant, and credit balance.
Actions run
Free, metered, blocked, failed, and replayed actions all hit the same ledger.
Success bills
Only successful billable line items debit credits and create revenue.
Cart closes
Credits are debited, idempotency is locked, and the period is ready to close.
Merchant gets paid
The merchant keeps the transaction revenue; Nitrograph keeps the commerce margin.
Evidence drops out
Finance gets usage export, invoice reference, webhook delivery, and audit trail.
webhook.billing_period.closed delivered in 283msThe moat is cross-merchant
agent reputation.
You only see your own agent traffic.
Any SaaS team can meter its own agents. Only Nitrograph sees behavior across many merchants, rails, actions, approvals, and billing outcomes.
↳ one merchant → partial signalReputation travels with the agent.
When an agent reaches your software, you should know whether it pays reliably, respects caps, retries safely, and avoids abusive patterns.
↳ cross-merchant → trust memoryThat is hard to build alone.
The more merchants meter through Nitrograph, the better the agent reputation graph becomes, and the safer every merchant launch gets.
↳ network effect → checkout moatDiscovery is the funnel that
brings agents to merchants.
// Keep the index. Demote it to the demand path: get listed, get found, get billed.
Agents find the right merchant workflow.
18,000+ services ranked by relevance, trust, operational memory, and payment rail. Discovery gets agents to the merchant surface where Checkout can meter the work.
The map lowers agent integration failure.
We run a probe fleet against every service in the catalog. Version drift, schema quirks, undocumented caps, param combinations that silently return empty. A better map means more successful billable actions.
The agent-native path into Checkout.
One install. Four tools. Every MCP-compatible client (Claude, Cursor, Cline, Zed) gets discovery, service maps, outcome reporting, and pattern reporting as first-class tool calls. The developer funnel stays live; the merchant funnel gets paid.
Your traffic never enters the graph.
Nitrograph is not a proxy. We process the discovery queries, task context, outcome diagnoses, and workflow reports your agent sends to Nitrograph. We do not automatically see the downstream payload sent to the selected service or the service's response.
Operational memory comes from our own probe fleet. We run structured tests against every service in the catalog, and agents can report integration outcomes. Aggregated reports become gotchas and proven patterns so the next agent avoids the same failure.
/ graph stores
- service_id
- endpoint_path
- probe_result
- latency_ms
- discovery_query
- outcome_diagnosis
- workflow_steps
- rail
- version_hash
/ graph never stores
- your_prompts
- your_api_keys
- your_user_data
- downstream_payloads
- service_responses
- private_keys
Is your software ready to bill agents?
Agent Checkout turns your product's agent surface from readable and callable into billable. The gap is not just API access. It is package rules, customer approval, metering, caps, revocation, and proof finance can trust.
Your website is for humans. Your docs are for developers. Your agent surface is for autonomous software buyers, operators, and customer-approved agents acting inside existing accounts with explicit spend authority.
$ POST /v1/headless/inquiries
RETRIEVE
+ website
+ docs / API reference
MAP
→ workflows, approvals, schemas, usage, trust
RECOMMEND
→ merchant path, audit, sprint, or monitoring
Billable
Package rules, customer approval, credit debits, billing evidence, and revocation make agent work commercially safe.
Readable
Make your product understandable to agents: services, workflows, schemas, permissions, and constraints.
Callable
Expose the workflows customers want agents to run through APIs, MCP, service catalogs, and call cards.
Payment rails are the plumbing,
not the pitch.
// Two rails live today. More as they ship.
x402
USDC on Base, Polygon, and Solana. Coinbase co-founded, open standard. Pay-per-call with receipts on-chain. 18,000+ services in the catalog.
// Catalog surfaces every x402 service's payment contract GET /v1/service/x402-lead-enrichment-api rail: "x402" cost: { amount: "0.001", currency: "USD" } // agent calls the service → handles its 402 → pays → 200 OK
MPP
Session-based streaming payments on Tempo. Sub-second settlement, delegated spend authority, no per-call signing. Co-designed by Stripe and Tempo. 90 services in the catalog.
// Catalog exposes every MPP service's session endpoint GET /v1/service/apollo rail: "mpp" cost: { amount: "0.0056", currency: "USDC" } // agent opens session with service → streams calls → settles
TAP
Identity-bound agent transactions with reputation carry-over. For services that require a known caller, not just a funded one.
// coming when spec stabilizes agent = nitro.tap.whoami() agent.rep.score → 0.94 · 12,403 tx
A2A
Agents paying agents directly. Nitrograph routes intent to the best counterparty, not just the best API.
// route intent → human-or-agent service nitro.route({ intent: "legal-review" }) ↳ matched agent://firm.delacroix/contracts
Built on HTTP. Works with
every agent framework.
Hosted MCP
https://api.nitrograph.com/mcpREST API
POST /v1/discoverGET /v1/service/:slugLocal MCP (stdio)
npx nitrographAgent Harness
npm i nitrographFramework adapters
langchain · crewai · vercel-ai · mastra · pydantic-aiTurn API calls into billable agent actions.
Price usage when agents create value.
Your API suite becomes an agent commerce surface: package the endpoints customers want automated, attach credits or per-action prices, and meter only approved successful calls. Nitrograph supplies the customer controls, usage ledger, and billing evidence; you keep the merchant relationship and pay through transaction margin.
Request a PilotWrap the endpoints agents need as priced actions, workflow packs, or credit-backed products.
Approved calls that complete create billable line items. Failed, denied, or out-of-scope requests stay visible at $0.
Close each cart with customer, grant, action, credits, and invoice references finance can audit.