Meta-MCP Wizard

The MCPHero Meta-MCP server lets agents build persistent tools without leaving their MCP client. Instead of re-explaining an API or re-generating SQL on every run, the agent creates a hosted MCP server once and calls it forever — saving 95-99% on token usage for recurring tasks.

Connect it to Claude Desktop, Cursor, or any MCP client, and your agent can build, deploy, and register new MCP servers on the fly.

Meta-MCP endpoint: https://api.mcphero.app/mcp/meta/mcp

Setup

Add the Meta-MCP server to your MCP client config:

Claude Desktop

// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "mcphero": {
      "url": "https://api.mcphero.app/mcp/meta/mcp"
    }
  }
}

Config file locations:

• macOS: ~/Library/Application Support/Claude/claudedesktopconfig.json
• Windows: %APPDATA%\Claude\claudedesktopconfig.json
• Linux: ~/.config/claude/claudedesktopconfig.json

Cursor and other MCP clients

Add the same URL https://api.mcphero.app/mcp/meta/mcp to your client's MCP server config. The client handles OAuth 2.1 authentication automatically on first use.

Available Tools

The Meta-MCP server exposes these tools:

Async vs Sync Tools

Async (trigger then poll wizardstate): wizardstart, wizardrefinetools, wizardsuggestenvvars, wizardrefineenvvars, wizardgeneratecode

Sync (wait for result): wizardcreatesession, wizardchat, wizardlisttools, wizardsubmittools, wizardlistenvvars, wizardsubmitenvvars, wizardsetauth, wizardregeneratetoolcode, wizarddeploy, wizardstate

The Wizard Pipeline

1. wizard_create_session    → Returns server_id (save it, needed everywhere)
2. wizard_chat (loop)       → Gather requirements; stop when is_ready: true
3. wizard_start             → Transition to tool suggestion (async → poll)
4. wizard_list_tools        → Review AI-suggested tools
5. wizard_refine_tools      → Iterate on tools until satisfied (optional, async → poll)
6. wizard_submit_tools      → Confirm selection by tool UUID list
7. wizard_list_env_vars     → Review suggested env vars
8. wizard_refine_env_vars   → Iterate on env vars (optional, async → poll)
9. wizard_submit_env_vars   → Provide values as {uuid: value} dict (call even if empty)
10. wizard_set_auth         → Generate bearer token
11. wizard_generate_code    → Trigger code generation (async → poll)
12. wizard_deploy           → Deploy → returns server_url + bearer_token

Always call wizardsubmitenvvars, even when wizardlistenvvars returns []. Pass an empty dict {} so the backend transitions to the next state.

State Machine

The setupstatus field in wizardstate tells you where you are:

gathering_requirements     → User is chatting about requirements
tools_generating           → LLM is generating tool suggestions (async, poll)
tools_selection            → Tools ready for review/selection
env_vars_generating        → LLM is generating env var suggestions (async, poll)
env_vars_setup             → Env vars ready for review/submission
auth_selection             → Ready for auth setup
code_generating            → LLM is generating code (async, poll)
code_gen                   → Code ready for review
deployment_selection       → Ready to deploy
ready                      → Server deployed and live

States ending in generating are transient — poll until they transition. The processingstatus field is the reliable check: "idle" means done, "processing" means wait, "error" means check processing_error.

Polling Pattern

After any async tool, poll wizardstate until processingstatus is "idle":

wizard_state(server_id) → check .processing_status
  "processing"  → wait and call again
  "idle"        → ready for next step
  "error"       → check .processing_error

Connecting a Deployed Server to MCP Clients

After wizard_deploy, construct the full server URL:

https://api.mcphero.app{server_url}

where server_url is the relative path returned (e.g., /mcp/<server-id>/mcp).

Claude Desktop config

{
  "mcpServers": {
    "my-server": {
      "url": "https://api.mcphero.app/mcp/<server-id>/mcp",
      "headers": {
        "Authorization": "Bearer <bearer_token>"
      }
    }
  }
}

Key Tips

Free tier: Max 5 tools per server. wizardsubmittools will error if more are selected.

serverid is everything: Save the UUID returned from wizardcreate_session. Every subsequent call needs it.

Empty env vars: Even if wizardlistenvvars returns [], you must still call wizardsubmitenvvars with {} so the backend transitions to the next state.

Regenerate without redeploy: After using wizardregeneratetool_code, the code change takes effect immediately for already-deployed servers (the server auto-remounts).