gatefareio/mcp-server
Marketplace MCP for paid HTTP APIs. Pay per call in USDC on Base via the open x402 standard — non-custodial. 13 tools for discovery, buying, and publishing APIs.
README
@gatefare/mcp
Give your AI agent a wallet and a marketplace.
@gatefare/mcpis a Model Context Protocol server that connects Claude Desktop, Cursor, or any MCP-compatible agent to the Gatefare catalog of paid HTTP APIs. Payments settle as USDC on Base via the open x402 standard — no SaaS keys, no subscriptions, no escrow. Non-custodial: signing happens locally; the private key never leaves your machine.

┌─────────────┐ ┌──────────────┐ ┌─────────────────┐
│ Claude / │ MCP stdio │ @gatefare/mcp│ HTTP + x402 │ gatefare.io │
│ Cursor / │ ─────────────► │ (this repo)│ ─────────────► │ proxy + │
│ your agent │ │ │ │ catalog │
└─────────────┘ └──────┬───────┘ └─────────────────┘
│
│ EIP-3009 sign
▼
┌─────────────┐
│ Base USDC │
└─────────────┘
Quick start
1. Drop into your client
Claude Desktop — ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"gatefare": {
"command": "npx",
"args": ["-y", "@gatefare/mcp"]
}
}
}
Cursor — ~/.cursor/mcp.json or project-level .cursor/mcp.json:
{
"mcpServers": {
"gatefare": {
"command": "npx",
"args": ["-y", "@gatefare/mcp"]
}
}
}
Restart the client. The agent now has 5 read-only tools — discovery + safety. Try:
"Search Gatefare for weather APIs."
2. Add a wallet to make paid calls
Add env to the same config:
{
"mcpServers": {
"gatefare": {
"command": "npx",
"args": ["-y", "@gatefare/mcp"],
"env": {
"WALLET_PRIVATE_KEY": "0xYOUR_KEY",
"WALLET_BUDGET_USD": "5.00"
}
}
}
}
Buyer tools (call_api, get_wallet_balance, estimate_cost) become available. The
WALLET_BUDGET_USD cap is a runtime safety net — for a hard cap, fund the wallet with
only what you're willing to spend.
"What's London's weather right now? Spend up to $0.001."
3. (Optional) Publish your own APIs
Get a PAT at gatefare.io/dashboard/tokens and add:
"env": {
"GATEFARE_PAT": "gfpat_..."
}
Publisher tools (register_api, list_my_apis, update_api, get_revenue,
distribute) appear.
"Publish my API at https://api.example.com/sentiment for $0.001 per call."
Tools
13 tools across 4 domains. Tools auto-register based on which env vars are set — the agent never sees a tool it can't use.
Discovery — always available
| Tool | Description |
|---|---|
gatefare.search_apis |
Full-text search the catalog with filters (price, category, sort) |
gatefare.get_api |
Full details for one API by slug or handle/urlName |
gatefare.list_categories |
All categories with API counts |
gatefare.suggest |
Autocomplete suggestions for a query string |
Buyer — needs WALLET_PRIVATE_KEY
| Tool | Description |
|---|---|
gatefare.call_api |
Make a paid call. Handles 402 → sign → retry automatically |
gatefare.get_wallet_balance |
USDC + ETH on Base, plus remaining runtime budget |
gatefare.estimate_cost |
Project total cost for N planned calls |
Publisher — needs GATEFARE_PAT
| Tool | Description |
|---|---|
gatefare.register_api |
Publish a new paid API |
gatefare.list_my_apis |
Your published APIs with stats |
gatefare.update_api |
Edit metadata, price, target URL |
gatefare.get_revenue |
Revenue time series + totals |
gatefare.distribute |
Trigger on-chain distribute() payout (destructive) |
Safety — always available
| Tool | Description |
|---|---|
gatefare.report_abuse |
Report a malicious / stolen API (DMCA, fraud, malware…) |
Configuration
| Var | Default | Required for |
|---|---|---|
GATEFARE_BASE_URL |
https://gatefare.io |
— (override for self-hosted) |
WALLET_PRIVATE_KEY |
— | Any buyer tool |
WALLET_BUDGET_USD |
unlimited | Optional spend cap |
WALLET_NETWORK |
eip155:8453 |
eip155:84532 for Sepolia testnet |
GATEFARE_PAT |
— | Any publisher tool |
LOG_LEVEL |
info |
debug for verbose stderr |
Examples
Discover & buy in one breath (Claude Desktop)
You: Find me a sub-$0.001 weather API and call it for "Tokyo".
Claude: Calling
gatefare.search_apiswithmax_price: 0.001…
Founddemo-weatherby @alice at $0.001/call.
Callinggatefare.call_apiwithslug: "demo-weather",query: {city: "Tokyo"}…
Tokyo is 22°C, partly cloudy. Paid 0.001 USDC. Receipt:settled-tx-0x9a…
Programmatic — Python agent
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client
server = StdioServerParameters(
command="npx",
args=["-y", "@gatefare/mcp"],
env={"WALLET_PRIVATE_KEY": "0x...", "WALLET_BUDGET_USD": "1.00"},
)
async with stdio_client(server) as (r, w):
async with ClientSession(r, w) as s:
await s.initialize()
result = await s.call_tool(
"gatefare.call_api",
arguments={"slug": "demo-weather", "query": {"city": "Tokyo"}},
)
print(result.content[0].text)
See examples/ for runnable variants: Claude Desktop, Cursor, Python,
TypeScript, and a pure-discovery walkthrough.
Not building an AI agent? Picking the right tool
If you want to pay for x402 APIs from a backend (no agent), use Coinbase's
official x402 SDKs — x402-python (PyPI), coinbase/x402/go, …/java, or
@x402/fetch. They handle the
payment flow; you don't need this MCP server.
If you want to browse the Gatefare catalog from any language, hit the
REST API directly: gatefare.io/api/catalog
(OpenAPI 3.1 spec).
Full breakdown of which tool fits which use case in
docs/integrations.md.
Direct CLI (for debugging)
# Run the server in foreground; talks JSON-RPC over stdio.
npx -y @gatefare/mcp
# In another terminal, send a frame:
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' | \
npx -y @gatefare/mcp
Errors
Tool results include isError: true and a structured body
{ error: <code>, message: <human>, details?: <any> }. Codes are stable —
agents can switch on them for retry / surfacing logic.
| Code | Meaning |
|---|---|
INVALID_INPUT |
Input failed zod validation |
WALLET_NOT_CONFIGURED |
Set WALLET_PRIVATE_KEY for buyer tools |
PAT_NOT_CONFIGURED |
Set GATEFARE_PAT for publisher tools |
BUDGET_EXHAUSTED |
Runtime budget cap hit |
INSUFFICIENT_BALANCE |
Wallet doesn't have enough USDC |
PRICE_TOO_HIGH |
Server's price exceeds your max_price |
API_NOT_FOUND |
Slug doesn't exist or is suspended |
UPSTREAM_ERROR |
Paid API returned non-2xx, or its 402 was malformed |
RATE_LIMITED |
Gatefare rate-limited the request |
NETWORK_ERROR |
Could not reach Gatefare |
GATEFARE_API_ERROR |
Gatefare returned a 4xx / 5xx |
How it works (the 30-second version)
- The agent calls
gatefare.call_api { slug: "demo-weather", … }. - We
GET https://gatefare.io/p/demo-weather(no payment yet). - The Gatefare proxy returns 402 Payment Required with
accepts: [{network, payTo, maxAmountRequired, …}]. - We sign an EIP-3009
transferWithAuthorizationfor that exact amount and recipient on the configured network. - We retry the request with the signed
X-Paymentheader (base64-encoded JSON, x402 v2). - Gatefare verifies the signature, settles the USDC transfer, and proxies the call to the upstream API.
- We hand the upstream response (and a payment receipt) back to the agent.
The signature is single-use, time-bounded, and never leaves your machine for any
purpose other than this exact transfer to this exact payTo. The private key
is never logged.
Security
- Non-custodial. Private keys live in your env, signing happens locally, no Gatefare service ever sees them.
- Network confusion-resistant. A malicious gateway returning Sepolia-only requirements to a mainnet user is rejected — we never sign for a chain the user didn't configure.
- Cryptographically random nonces. No
Date.now()-based collisions. - Validity window clamped to 1 hour even if the server requests more.
- Strict input validation. Slugs are
^[a-z0-9_-]+$and URL-encoded; no path traversal.targetUrlblocksfile://,localhost, cloud metadata IPs,.local, and.internalhosts at registration time. - Secret hygiene. Tests assert that the private key and PAT never appear in stderr / stdout, ever.
Development
git clone https://github.com/gatefareio/mcp-server.git
cd mcp-server
npm install
npm run typecheck
npm test # 138 unit tests
npm run test:e2e # 10 e2e tests against live gatefare.io (set GATEFARE_E2E=1)
npm run build
To use a local checkout in your client config:
npm link
# in claude_desktop_config.json:
# "command": "gatefare-mcp"
Architecture
src/
├── index.ts # entry — wires stdio transport
├── server.ts # McpServer instance + tool registration
├── config.ts # env parsing, capability detection
├── client.ts # REST client (wraps fetch)
├── x402.ts # 402 parsing + EIP-3009 signing
├── types.ts # shared types + GatefareError
└── tools/
├── discovery.ts # search_apis, get_api, list_categories, suggest
├── buyer.ts # call_api, get_wallet_balance, estimate_cost
├── publisher.ts # register_api, list_my_apis, update_api, get_revenue, distribute
└── safety.ts # report_abuse
Test layout
tests/
├── config.test.ts # env parsing edges
├── client.test.ts # HTTP client error mapping
├── x402.test.ts # signing + parsing primitives
├── x402-flow.test.ts # full 402 → sign → retry handshake (mocked fetch)
├── server.test.ts # capability-driven tool registration
├── init.test.ts # subprocess: bootstrap, env crashes, secret leakage
├── stdio-protocol.test.ts # stdout pollution + recovery from tool errors
├── stability.test.ts # 100 concurrent calls, memory baseline, ReDoS
├── tools/
│ ├── discovery.test.ts
│ ├── buyer.test.ts
│ ├── buyer-flow.test.ts
│ ├── publisher.test.ts
│ └── safety.test.ts
└── integration/
└── e2e.test.ts # real gatefare.io, gated by GATEFARE_E2E=1
Contributing
Issues and PRs welcome. See CONTRIBUTING.md for the workflow, style guide, and how to add a new tool.
License
MIT © Gatefare
Links
- 🌐 Marketplace: gatefare.io
- 📚 API docs: gatefare.io/docs
- 🤖 LLM context (single file): gatefare.io/llms-full.txt
- 📐 OpenAPI spec: gatefare.io/openapi.json
- 🐦 Twitter: @Gatefareio
- 🔌 Model Context Protocol: modelcontextprotocol.io
- 💸 x402 standard: x402.org
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。