Create MCP Servers using Meta-MCP — 创建 MCP Servers using Meta-MCP

Meta-MCP Wizard

The MCPHero Meta-MCP server lets 代理s build persistent 工具s without leaving their MCP 命令行工具ent. Instead of re-explAIning an API or re-generating SQL on every 运行, the 代理 创建s a hosted MCP server once and calls it forever — saving 95-99% on 令牌 usage for recurring tasks.

Connect it to Claude 桌面, Cursor, or any MCP 命令行工具ent, and your 代理 can build, 部署, and register new MCP servers on the fly.

Meta-MCP 端点: https://API.mcphero.应用/mcp/meta/mcp

设置up

添加 the Meta-MCP server to your MCP 命令行工具ent config:

Claude 桌面 // ~/库/应用 Support/Claude/claude_桌面_config.json { "mcpServers": { "mcphero": { "url": "https://API.mcphero.应用/mcp/meta/mcp" } } }

Config file locations:

macOS: ~/库/应用 Support/Claude/claude_桌面_config.json Windows: %应用DATA%\Claude\claude_桌面_config.json Linux: ~/.config/claude/claude_桌面_config.json Cursor and other MCP 命令行工具ents

添加 the same URL https://API.mcphero.应用/mcp/meta/mcp to your 命令行工具ent's MCP server config. The 命令行工具ent handles OAuth 2.1 authentication automatically on first use.

AvAIlable 工具s

The Meta-MCP server exposes these 工具s:

工具 Purpose wizard_创建_会话 启动 会话; returns server_id wizard_chat 发送 requirement message; returns is_ready flag wizard_启动 Transition to 工具 suggestion (a同步) wizard_列出_工具s 列出 suggested 工具s with IDs, code, params wizard_refine_工具s Refine with feedback (a同步) wizard_submit_工具s Confirm selection by 工具 UUID 列出 wizard_suggest_env_vars Re-trigger env var suggestion (a同步) wizard_列出_env_vars 列出 env vars with IDs and descriptions wizard_refine_env_vars Refine env vars with feedback (a同步) wizard_submit_env_vars Submit values as {uuid: value} dict wizard_设置_auth 生成 bearer 令牌 wizard_生成_code Trigger code gen (a同步) wizard_re生成_工具_code Re生成 single 工具 code (同步) wizard_部署 部署 and 获取 server URL wizard_状态 Poll current 状态 — use after all a同步 steps A同步 vs 同步 工具s

A同步 (trigger then poll wizard_状态): wizard_启动, wizard_refine_工具s, wizard_suggest_env_vars, wizard_refine_env_vars, wizard_生成_code

同步 (wAIt for 结果): wizard_创建_会话, wizard_chat, wizard_列出_工具s, wizard_submit_工具s, wizard_列出_env_vars, wizard_submit_env_vars, wizard_设置_auth, wizard_re生成_工具_code, wizard_部署, wizard_状态

The Wizard 流水线

• wizard_创建_会话 → Returns server_id (save it, needed everywhere)
• wizard_chat (loop) → Gather requirements; 停止 when is_ready: true
• wizard_启动 → Transition to 工具 suggestion (a同步 → poll)
• wizard_列出_工具s → Review AI-suggested 工具s
• wizard_refine_工具s → Iterate on 工具s until satisfied (optional, a同步 → poll)
• wizard_submit_工具s → Confirm selection by 工具 UUID 列出
• wizard_列出_env_vars → Review suggested env vars
• wizard_refine_env_vars → Iterate on env vars (optional, a同步 → poll)
• wizard_submit_env_vars → Provide values as {uuid: value} dict (call even if empty)
• wizard_设置_auth → 生成 bearer 令牌
• wizard_生成_code → Trigger code generation (a同步 → poll)
• wizard_部署 → 部署 → returns server_url + bearer_令牌

Always call wizard_submit_env_vars, even when wizard_列出_env_vars returns []. Pass an empty dict {} so the backend transitions to the next 状态.

状态 Machine

The 设置up_状态 field in wizard_状态 tells you where you are:

gathering_requirements → User is chatting about requirements 工具s_generating → LLM is generating 工具 suggestions (a同步, poll) 工具s_selection → 工具s ready for review/selection env_vars_generating → LLM is generating env var suggestions (a同步, poll) env_vars_设置up → Env vars ready for review/submission auth_selection → Ready for auth 设置up code_generating → LLM is generating code (a同步, poll) code_gen → Code ready for review 部署ment_selection → Ready to 部署 ready → Server 部署ed and live

状态s ending in _generating are transient — poll until they transition. The processing_状态 field is the reliable 检查: "idle" means done, "processing" means wAIt, "error" means 检查 processing_error.

Polling Pattern

After any a同步 工具, poll wizard_状态 until processing_状态 is "idle":

wizard_状态(server_id) → 检查 .processing_状态 "processing" → wAIt and call agAIn "idle" → ready for next step "error" → 检查 .processing_error

Connecting a 部署ed Server to MCP 命令行工具ents

After wizard_部署, construct the full server URL:

https://API.mcphero.应用{server_url}

where server_url is the relative path returned (e.g., /mcp//mcp).

Claude 桌面 config { "mcpServers": { "my-server": { "url": "https://API.mcphero.应用/mcp//mcp", "headers": { "Authorization": "Bearer "