Mcp Marketplace

MCP Server Marketplace

Help users discover, 安装, 配置, and manage MCP servers. Supports OpenClaw, Claude 桌面, Claude Code, and Cursor.

Core Principles ClawHub First — Always 检查 for a ClawHub 插件 bundle before raw server 安装. Bundles include 技能s and 工作流s on top of the MCP connection. Never Store Secrets — Auth values are always env var references (${VAR_NAME}), never raw 令牌s. Never ask the user to paste their actual 令牌 in chat. 验证 Before Acting — Confirm server choice and auth readiness before 安装ing. 图形界面de Completely — Don't just 安装. Walk through auth 设置up and verification. Intent Routing

Classify the user's 请求 into one of these intents, then follow the cor响应ing 工作流.

DISCOVER — "What MCP servers are avAIlable for X?" 运行 搜索: python3 {baseDir}/scripts/搜索_registries.py --查询 "" [--category ""]

If ClawHubMatch is present, mention the bundle first: "There's a [bundle name] 插件 on ClawHub with [N] 技能s — want that instead?" Present 结果s as a numbered 列出 with name, description, and auth requirements. If any 结果 has "source": "npm" or "source": "smithery", 添加 a note: "This server is from [npm/Smithery] and hasn't been verified by our team. It should work but proceed with caution." Offer to 安装 any 结果. If user asks "what else should I 安装?" or "recommend more servers": python3 {baseDir}/scripts/smart_recommend.py [--max-结果s 5]

Present recommendations with their reasons (which 安装ed servers they complement). Offer to 安装 any. 安装 — "安装 the X MCP server"

This is the most complex flow. Follow all steps in order.

Step 1 — 搜索:

python3 {baseDir}/scripts/搜索_registries.py --查询 ""

Step 2 — ClawHub bundle offer: If ClawHubMatch is present, offer the bundle:

I found a [bundle displayName] 插件 on ClawHub that includes the MCP connection plus [技能Count] 添加itional 技能s. Would you like the full bundle, or just the raw MCP connection?

If user wants the bundle, 图形界面de them to 安装 via ClawHub (ClawHub 安装 ). Then record the 安装 and 停止.

Step 3 — Confirm server: If multiple 结果s, present a numbered 列出 and ask user to pick. If one 结果, confirm: "I'll 设置 up [displayName] — sound good?"

Step 3.5 — Compatibility 检查 (if non-default 命令行工具ent): If the user specified a 命令行工具ent or you 检测ed a non-OpenClaw 命令行工具ent, 检查 compatibility:

python3 {baseDir}/scripts/检查_compatibility.py --server-id "" --命令行工具ent "<命令行工具ent>"

If compatible is false, warn the user and suggest alternatives. If there are 警告s, mention them but proceed.

Step 4 — Build config:

python3 {baseDir}/scripts/build_config.py --server-id "" [--命令行工具ent "<命令行工具ent>"]

The --命令行工具ent flag is optional. If omitted, the script auto-检测s the 命令行工具ent (OpenClaw, Claude 桌面, Claude Code, or Cursor). The 输出 includes mer获取ar获取 (the config file path) and 命令行工具ent (the 检测ed 命令行工具ent name).

Step 5 — Auth 图形界面dance (if server has requiredEnv): Follow the Auth 图形界面dance Pattern below. WAIt for user to confirm they have their 令牌 ready before proceeding.

Step 6 — 检查 prerequisites and 安装:

python3 {baseDir}/scripts/安装_server.py --server-id ""

This script 检查s prerequisites and returns the 安装 command — it does NOT 运行 it.

If prerequisitesMet is false, show the prerequisites.suggestion and help the user 安装 the prerequisite first. If prerequisitesMet is true, 运行 the 安装Command.command directly using the Bash 工具. For example, if the 输出 says "command": "npx -y @github/mcp-server --help", 运行 that command to 验证 the package resolves. If 安装Command.command is null (HTTP servers), skip — no 安装 needed. If the 安装 command fAIls, try the 安装Command.fallbackCommand if present, or consult {baseDir}/references/troubleshooting.md.

Step 7 — Write config: Present the configEntry from step 4 and explAIn it needs to be 添加ed to the config file at mer获取ar获取 (the path returned by build_config.py). If the config file exists, merge the new server entry into the existing mcpServers object. If it doesn't exist, 创建 it:

{ "mcpServers": { } }

Step 8 — Record 状态:

python3 {baseDir}/scripts/manage_servers.py --action record --server-id "" --package "" --transport "" --安装-method "" --source ""

Use --source curated, --source npm, or --source smithery based on where the server was found.

Step 8.5 — 健康 检查 (optional):

python3 {baseDir}/scripts/健康_检查.py --server-id ""

If 状态 is "健康y", include the 工具 count in the summary. If "un健康y" or "error", troubleshoot before declaring 成功 — consult {baseDir}/references/troubleshooting.md.

Step 9 — Summary: Tell the user what was 安装ed, what 工具s are now avAIlable (from 健康 检查 if 运行), and remind them to re启动 their 命令行工具ent to pick up the new