mcpField
H
HTTP 402 AI Tollbooth

HTTP 402 AI Tollbooth

Remote
github.com/RobotFleet-HQDeveloper Tools0Updated 7w ago
Repository

About HTTP 402 AI Tollbooth

HTTP 402 AI Tollbooth provides developers and AI agents with structured troubleshooting datasets, offering three records per product as a free preview and full datasets through Bitcoin Lightning payment. Its MCP tools support catalog discovery, previews, payment-method information, and purchase instructions; purchases remain human-directed.

Key features

  1. 01Browse a catalog of troubleshooting datasets
  2. 02Preview three records per product
  3. 03Unlock full datasets via Lightning payment
  4. 04Use stable fields for errors and fixes
  5. 05Replay protection for redeemed datasets

Use cases

  1. 01Preview fixes for a developer error
  2. 02Find datasets for an agent troubleshooting flow
  3. 03Review purchase instructions before paying

MCP support

A live Streamable HTTP JSON-RPC endpoint exposes four read-only tools for product listings, previews, payment methods, and purchase discovery. The purchase tool returns payment instructions and an HTTP 402 challenge; it does not pay or purchase on the caller's behalf.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://http-402-ai-tollbooth.rsaun-lightning.workers.dev/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

Tools

Details

From the project README

HTTP 402 AI Tollbooth

Agent-ready developer troubleshooting data, sold per request over Bitcoin Lightning.

Live: https://http-402-ai-tollbooth.rsaun-lightning.workers.dev

HTTP 402 AI Tollbooth gives developers and AI agents three useful records free, then unlocks complete structured datasets with a BOLT11 Lightning payment. There are no accounts, subscriptions, credit cards, or API keys.

This is a public, documentation and example repository. The production Worker implementation remains private and is not included here.

Try it now

List the complete catalog:

curl -sS https://http-402-ai-tollbooth.rsaun-lightning.workers.dev/api/catalog

Preview exactly three records without payment:

curl -sS https://http-402-ai-tollbooth.rsaun-lightning.workers.dev/api/preview/api-error-fixes

Request the full product and receive an HTTP 402 Lightning challenge:

curl -i https://http-402-ai-tollbooth.rsaun-lightning.workers.dev/api/data/api-error-fixes

The last command creates an invoice. Do not run it unless you intentionally want a payment challenge; simply requesting it does not spend sats.

Why agents use it

  • 203 products across DevOps, APIs, ML/AI, Web/Frontend, and Databases
  • Entry, Standard, and Premium datasets with 50, 100, or 200 records
  • Exactly three free records per product for try-before-you-buy discovery
  • Stable error_code, description, and programmatic_fix fields
  • HTTP 402 and Bitcoin Lightning payments without user accounts
  • OpenAPI 3.1, MCP discovery metadata, and llms.txt
  • Deterministic fallback generation for resilient dataset delivery
  • Durable one-time redemption and replay protection
  • Future x402 settlement is documented but remains disabled pending verification

Architecture

AI agent / developer
        |
        | HTTPS: catalog, preview, paid dataset
        v
Cloudflare Worker
   |         |                    |
   |         |                    +--> Alby --> Bitcoin Lightning invoice
   |         |
   |         +--> Durable Objects --> atomic redemption + counters
   |
   +--> Workers KV --> invoice challenges + generated dataset cache

Pricing

TierPriceRecords
Entry2,000 sats50
Standard5,000 sats100
Premium10,000 sats200

See the pricing rationale, including the live-verified 2,000-sat receiving floor.

How to pay

  1. Request a paid product endpoint.
  2. Receive 402 Payment Required with a BOLT11 invoice and payment hash.
  3. Pay the invoice in a Lightning wallet.
  4. Retry the same endpoint with X-Payment-Hash and X-Payment-Preimage.
  5. Receive the full dataset once; replayed proofs are rejected.

Never publish or share a payment preimage before using it for its intended redemption. Future Solana USDC/x402 support is disabled until its manual verification process is completed.

Discovery documents

Integrations

  • MCP metadata: tool descriptions for catalog, preview, and purchase discovery
  • MCP transport: a live Streamable HTTP JSON-RPC endpoint at /mcp with four read-only tools
  • OpenAPI: complete HTTP endpoint descriptions for agent frameworks
  • LLMs.txt: concise machine-readable project and payment guidance

Connect an MCP client to https://http-402-ai-tollbooth.rsaun-lightning.workers.dev/mcp. The server supports list_products, preview_product, list_payment_methods, and buy_product. The last tool only returns payment directions and an HTTP 402 challenge; it never pays an invoice or creates a purchase on the caller's behalf.

For a complete no-payment example, see the Node agent recovery sample.

Runnable free sample

The sample shows how an agent can inspect malformed JSON, retry a transient 429 once after Retry-After, and safely stop when a response is not recoverable. It uses Node 22+ built-ins only:

cd examples/node-agent-recovery
npm test
npm start

The demo calls only the public three-record preview endpoint. It does not call a paid route, create an invoice, or automatically replay a payment proof.

License

The documentation in this billboard repository is available under the MIT License. The private production source code is not covered by this repository or license.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.