StackResolve

dev.stackresolve由 stackresolve.dev 发布官方发布者v0.4.0更新于 Oct 24, 2026

Find, compare, and audit software for AI agents. Scored registry of tools and MCP servers.

已验证Streamable HTTP可网页运行网页搜索与抓取开发者工具商业与电商

概览

AI 生成的概览

远程 MCP 服务器,用于为 AI 代理查找、比较和审计软件与 MCP 服务器,并提供结构化的公司调研。

可以做什么
  • 把自然语言任务解析为按匹配度排序的、适合代理使用的工具,可加需求筛选并联网发现
  • 搜索、列出和比较注册表中的产品,读取含 AgentReady 评分的完整档案
  • 获取带引用的最新做法指南,并审计产品的代理就绪度
  • 根据 OpenAPI 生成代理接口:MCP 服务器、CLI、llms.txt 和安装片段
  • 查询结构化公司记录、标准化定价、竞争对手及公司横向对比
  • 查看当前计费周期的自身用量
适用场景
当助手需要为某项任务挑选软件或 MCP 服务器、判断产品是否适合代理使用,或需要带引用的公司信息与定价事实而不想自己抓取网页时适用。
运行要求
远程 streamable HTTP 端点,无需安装。注册表读取无需密钥;resolve、how_to、audit、research 等计费工具需要免费 API 密钥,通过 x-api-key 请求头发送。可选的 npm CLI 和 SDK 需要 Node 18 或更高版本。
安装前请注意
x-api-key 请求头属于机密,不要写入共享配置。计费工具会消耗密钥配额,get_usage 会显示计费周期用量。安装命令会生成密钥并修改本机的代理配置文件和技能文件,建议先用 dry-run 查看改动。请求内容和被审计的域名会发送给第三方服务。

安装

在 SourceWeft 中

  1. 打开 控制台中的 StackResolve,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/dev.stackresolve%2Fstackresolve/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "stackresolve": {
      "type": "http",
      "url": "https://mcp.stackresolve.dev/mcp"
    }
  }
}

README

stackresolve

Official TypeScript/Node SDK for the StackResolve API.

StackResolve helps agents discover, evaluate, select, install, and use software (AgentReady), and compresses web research into structured company data calls (CompanyData). This SDK is a thin, typed wrapper over the live REST API.

One line for your coding agent

bash
npx stackresolve@latest install

That mints an API key if you have none, registers the hosted MCP server for every coding agent on the machine (Claude Code, Cursor, Codex, Windsurf, VS Code), writes the StackResolve skill to ~/.claude/skills/stackresolve/SKILL.md, and verifies the key against the API. It is safe to re-run, and it backs up every file it changes.

bash
npx stackresolve@latest install --dry-run     # show what would changenpx stackresolve@latest install --cursor      # one agent onlynpx stackresolve@latest install --project --rules   # also wire up the current repo

MCP server

The hosted server needs no install. The handshake and the registry reads (search_tools, get_profile, compare_products, list_registry) work without a key. Metered tools (resolve, how_to, audit, research_company, get_company, get_pricing) need a free key from https://stackresolve.dev/developers, sent as the x-api-key header.

Claude Code:

bash
claude mcp add --transport http stackresolve https://mcp.stackresolve.dev/mcp

Cursor (~/.cursor/mcp.json):

json
{  "mcpServers": {    "stackresolve": {      "url": "https://mcp.stackresolve.dev/mcp",      "headers": { "x-api-key": "ar_..." }    }  }}

Codex (~/.codex/config.toml):

toml
[mcp_servers.stackresolve]url = "https://mcp.stackresolve.dev/mcp"
[mcp_servers.stackresolve.headers]x-api-key = "ar_..."

Tools: resolve, how_to, find_tools_for_task, search_tools, compare_products, get_profile, list_registry, audit, generate_agent_interfaces, get_company, get_pricing, research_company, find_competitors, compare_companies, get_usage.

CLI

bash
npm i -g stackresolvestackresolve find "scrape javascript-heavy sites from claude code"stackresolve audit stripe.comstackresolve company vercel.comstackresolve --help

Install

bash
npm install stackresolve      # SDKnpm i -g stackresolve         # CLI

Requires Node 18+ (uses the built-in global fetch). No runtime dependencies.

Quickstart

ts
import { StackResolve } from 'stackresolve'
const sr = new StackResolve({ apiKey: process.env.STACKRESOLVE_API_KEY })
// 1. Audit a domain: AgentReady scores, facts, and issues.const report = await sr.audit('stripe.com')console.log(report.scores.agentready, report.issues)
// 2. Structured company data.const company = await sr.getCompany('vercel.com')console.log(company.name, company.description)
// 3. Task -> ranked, agent-ready tools.const tools = await sr.findTools('send transactional email from a Node service')console.log(tools.results)
// 4. Run an Agent Discovery check (needs an API key).const discovery = await sr.runDiscovery('firecrawl')console.log(discovery)

Authentication

Get an API key at stackresolve.dev. Pass it to the constructor, or set the STACKRESOLVE_API_KEY environment variable:

ts
const sr = new StackResolve({ apiKey: 'ar_...' })// or, reading STACKRESOLVE_API_KEY from the environment:const sr2 = new StackResolve()

The key is sent as the x-api-key header on every request. Public endpoints (audit, search, profiles, company data, ...) work without a key, subject to an anonymous rate limit. Gated endpoints (monitors, usage, discovery runs) need one.

You can point the client at a different host with baseUrl or the STACKRESOLVE_BASE_URL environment variable:

ts
const sr = new StackResolve({ apiKey, baseUrl: 'https://api.stackresolve.dev' })

Errors

Any non-2xx response throws a StackResolveError carrying the HTTP status and the parsed body:

ts
import { StackResolve, StackResolveError } from 'stackresolve'
try {  await sr.getUsage()} catch (err) {  if (err instanceof StackResolveError) {    console.error(err.status, err.body)  }}

Methods

AgentReady: audit(domain), findTools(task), search(query, requirements?), compare(slugs), getProfile(slug), listRegistry(opts?), getCategories(), getCategory(slug), getDiscovery(slug), runDiscovery(slug), getScoreHistory(slug), generate(domain, openapiUrl?), deploy(domain, openapiUrl?).

CompanyData: getCompany(domain), getPricing(domain), getCompetitors(domain), research(domain, question?), compareCompanies(domains).

Monitors + account (API key required): listMonitors(), addMonitor(slug, opts?), runMonitorNow(slug), getUsage(), getMyProfile().

Other languages and frameworks

Python SDK:

bash
pip install stackresolve

LangChain tools, so an agent can pick software and research companies without a search-and-scrape loop:

bash
pip install langchain-stackresolve
python
from langchain_stackresolve import StackResolveToolkittools = StackResolveToolkit().get_tools()

REST directly: https://api.stackresolve.dev, OpenAPI at /openapi.json, llms.txt at /llms.txt.

License

MIT

来源:README.md,提交 3b3a585

工具

15

resolve

Best tools for a natural-language task: resolves the task to the capabilities it needs, returns products that expose them ranked by fit and AgentReady score, each with a reason. Optional requirements (api/mcp/self_serve/openapi/cli) filter candidates. Set discover:true to search the web and audit new products into the registry when coverage is thin (slower, costs more).

输入参数apiclimcptask必填openapidiscoverself_serve

how_to

BEFORE building something, call this: returns the CURRENT (date-stamped), cited, tool-aware way to do a task, so you build from today's best practice instead of training-cutoff memory. Composes live web research + the tool resolver + one synthesis pass into { approach, recommendedTools, steps, gotchas, deprecated, sources, asOf }.

输入参数task必填

find_tools_for_task

Rank the software an agent should use for a task.

输入参数task必填

search_tools

Search the registry by query + requirements (api/mcp/self_serve/openapi/cli).

输入参数apiclimcpquery必填openapiself_serve

compare_products

Compare products by slug.

输入参数slugs必填

get_profile

Full registry profile (AgentReady Score, capabilities, sources).

输入参数slug必填

list_registry

List the registry, highest AgentReady Score first.

输入参数limitcategoryminScore

audit

Audit a product for agent-readiness.

输入参数domain必填

generate_agent_interfaces

Make a product agent-native: from its OpenAPI, generate an MCP server (one tool per operation), a CLI, an llms.txt, and Claude Code/Codex/Cursor install snippets.

输入参数domain必填openapiUrl

get_company

Structured company record.

输入参数domain必填

get_pricing

Normalized pricing.

输入参数domain必填

research_company

Grounded, cited answer about a company.

输入参数domain必填question

find_competitors

Real competitors and alternatives for a company, extracted from live research and returned as structured rows.

输入参数limitdomain必填

compare_companies

Structured side-by-side of companies across fields (description, products, categories, funding, technology).

输入参数fieldsdomains必填

get_usage

Your usage this billing period (read-only).

无需输入参数。

版本历史

1
  1. v0.4.0最新Sep 16, 2026

评分

已满足适用检查项的 100%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分