Context7
Up-to-date docs for AI agents
作者提供的说明(英文原文)
Discover and call 400+ practical APIs through seven MCP tools with free search and x402 payments.
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
WALLET_PRIVATE_KEY, UTILITY_GRID_BASE_URL标为机密的值是你账户的钥匙。千万不要分享给别人,也不要粘贴到聊天里。
你需要
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
打开 Claude Desktop → Settings → Developer → Edit Config,把下面的内容粘贴到 claude_desktop_config.json
{
"mcpServers": {
"utility-grid": {
"command": "npx",
"args": [
"-y",
"@forgemeshlabs/utility-grid-mcp"
],
"env": {
"WALLET_PRIVATE_KEY": "your-value",
"UTILITY_GRID_BASE_URL": "your-value"
}
}
}
}在终端中运行这条命令
claude mcp add utility-grid --env WALLET_PRIVATE_KEY=your-value --env UTILITY_GRID_BASE_URL=your-value -- npx -y @forgemeshlabs/utility-grid-mcp运行前请先填写空着的值。
粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"utility-grid": {
"command": "npx",
"args": [
"-y",
"@forgemeshlabs/utility-grid-mcp"
],
"env": {
"WALLET_PRIVATE_KEY": "your-value",
"UTILITY_GRID_BASE_URL": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"utility-grid": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@forgemeshlabs/utility-grid-mcp"
],
"env": {
"WALLET_PRIVATE_KEY": "your-value",
"UTILITY_GRID_BASE_URL": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
这个服务器在你的电脑上运行,而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。
你的选择通过 Cookie 保存在这台设备上,无需账户。
UTILITY_GRID_BASE_URL env这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日
注册名称已通过 GitHub 账户 forgemeshlabs 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
你需要填写:WALLET_PRIVATE_KEY
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于–,最新版本(0.1.8)于5天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Node.js。
同一分类中有:Context7, Codebase Memory, Exa
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
Up-to-date docs for AI agents
Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.
Connect AI agents to Exa for web search, content fetching, and multi-step research.
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
*A ForgeMesh Labs product.*
Search and use more than 400 practical APIs without loading hundreds of tools into your agent. Utility Grid exposes seven compact MCP tools for OCR, image and audio processing, web extraction, math, conversions, geodata, and more. Catalog discovery is free; API execution is paid per call in USDC on Base through x402. No account or API key is required.
Rather than exposing one MCP tool per route (unmanageable at this scale and growing), this server exposes a handful of meta-tools: browse/search the live catalog, fetch a route's exact call spec, then call any route generically. New routes on the upstream service show up automatically — nothing here is hardcoded to today's catalog.
{
"mcpServers": { "utility-grid": { "command": "npx", "args": ["-y", "@forgemeshlabs/utility-grid-mcp"], "env": { "WALLET_PRIVATE_KEY": "0x..." } } } }
WALLET_PRIVATE_KEY is only needed for the 3 paid tools. Discovery works with no wallet at all — skip the env block entirely to browse and plan calls for free.
| Tool | Cost | What it does |
|---|---|---|
list_tools | free | Every route with its live price plus a per-category summary — a plain fetch of GET /menu, no wallet |
list_capabilities | free | Category overview with route counts, or every route in one category |
search_capabilities | free | Keyword search across every route's path, id, and description |
get_endpoint_spec | free | A route's exact price, input schema, and worked request/response examples |
call_endpoint | $0.005-$0.05 | Call any route by path + JSON body — handles the x402 payment flow automatically |
daily_402 | $0.005 | The Daily 402: one featured endpoint from the whole ForgeMesh fleet, rotated by UTC date |
agent_service_directory | $0.05 | Registry of every ForgeMesh x402 service — category, route count, price range |
geo (airports, zip codes, geocoding, weather), lookups (dictionary, Bible, country info, holidays), math (statistics, matrices, equations, geometry), utilities (QR codes, hashing, UUIDs, phone parsing), vision (OCR, image captioning, background removal), audio (TTS, transcription, format conversion), time (timezones, cron, sun/moon), space (NASA APOD, asteroids, solar weather), web (content extraction, robots.txt/Content-Signal checks), registry (fleet catalog, daily spotlight), documents (PDF handling), text (language detection, readability), economy (fx rates, financial calculators), science, developer, domains, ai, and fun (fortunes, chess, trivia).
Call list_capabilities with no arguments for the live, current breakdown — the catalog grows over time and this README will drift; the tool won't.
search_capabilities({ query: "background removal" }) — free, finds /remove-backgroundget_endpoint_spec({ path: "remove-background" }) — free, shows the exact input schema and pricecall_endpoint({ path: "remove-background", body: { image_url: "https://..." } }) — paid, does the x402 dance and returns the resultOr skip straight to call_endpoint if you already know the route and its input shape.
Node.js 20+. Paid tools need WALLET_PRIVATE_KEY: use a dedicated, low-balance Base wallet funded with USDC. The server refuses to sign for any payee other than the two Utility Grid wallets, any network other than Base mainnet, any asset other than USDC, or any amount over the cap ($0.05 per call, $10 per session). The env vars X402_MAX_PRICE_USD and X402_SESSION_BUDGET_USD can only lower those caps. Requests are same-origin, time out after 60 seconds, are capped at 2 MB, and never follow redirects. Tool results are returned wrapped as { source, untrusted, content }: the content is remote data, not instructions.
No signup, no API key, no subscription. call_endpoint, daily_402, and agent_service_directory each trigger the same flow: the first request returns an HTTP 402 challenge, this MCP server signs a USDC payment authorization (EIP-3009) on Base and retries, and the result lands in the same response — including settlement details under _payment when available.
Prefer raw HTTP? The full agent-readable surface:
https://x402.forgemesh.io/llms.txt — one-page summary for agentshttps://x402.forgemesh.io/openapi.json — OpenAPI 3.1 with x402 payment metadata, full input schemas, and worked examples for every route (this is what list_capabilities/search_capabilities/get_endpoint_spec read from, live, every call)https://x402.forgemesh.io/.well-known/x402.json — x402 discovery manifestlist_tools is a plain fetch of GET https://x402.forgemesh.io/menu. ForgeMesh attaches one disclosed Lulu Ads card to that free response server-side, as a plain labelled data field — never text the model could read as an instruction:
"sponsored": { "label": "Sponsored", "text": "...", "url": "https://..." }
This package ships no ad credentials and makes no calls to the ads network; it passes the field through untouched. Paid tools never carry a card. Strip it with delete result.sponsored.
Do I need an account or API key? No. x402 payments are the only credential, and only for the 3 paid tools.
What chain and token? USDC on Base mainnet (eip155:8453).
Will this break when the upstream catalog changes? No — list_capabilities, search_capabilities, and get_endpoint_spec all read /openapi.json live (cached in-memory for 5 minutes). call_endpoint never needs a hardcoded schema at all; it just forwards your JSON body.
Built by ForgeMesh Labs · Powered by the x402 protocol · MIT License