Quick Start

Get from zero to calling MCP tools in under 2 minutes.

Option A: Import Existing Config

If you already have MCP servers configured (.mcp.json from Claude Code, etc.):

mcpx init

Detected .mcp.json
Imported 3 servers:
  serena          uvx (daemon)
  seq-thinking    npx
  filesystem      npx
Created .mcpx/config.yml

Done. Skip to Use It.

Option B: Manual Config

Create .mcpx/config.yml in your project root:

servers:
  serena:
    command: uvx
    args:
      - --from
      - git+https://github.com/oraios/serena
      - serena
      - start-mcp-server
      - --project
      - "$(mcpx.project_root)"
    transport: stdio
    daemon: true

Or add a global server at ~/.mcpx/config.yml:

servers:
  sequential-thinking:
    command: npx
    args: ["-y", "@modelcontextprotocol/server-sequential-thinking"]
    transport: stdio

Use It

Discover servers

mcpx list
#   serena              uvx (daemon)
#   sequential-thinking npx

Discover tools

mcpx list serena -v

Shows every tool with all its flags — full discovery in one call.

Call a tool

mcpx serena search_symbol --name "UserAuth"

Get help for a tool

mcpx serena search_symbol --help

Shows flag names, types, required markers, descriptions, and defaults — all auto-generated from the MCP schema.

Inspect server capabilities

mcpx serena info
# Shows server name, version, protocol, and supported capabilities

Explore prompts and resources

If the server supports them:

mcpx <server> prompt list                       # list prompts
mcpx <server> prompt <name> --arg value         # get a prompt
mcpx <server> resource list                     # list resources
mcpx <server> resource read <uri>               # read a resource

Health check

mcpx ping serena
# serena: ok (21 tools, 47ms)

JSON output

mcpx serena search_symbol --name "Auth" --json

Pipe and compose

mcpx serena search_symbol --name "Auth" | jq -r '.[].file' | xargs code

What Just Happened

1. mcpx read your config to find the server
2. For daemon servers: spawned the server once, connected via unix socket
3. For direct servers: spawned a subprocess, did the MCP handshake
4. Translated your CLI flags into a JSON-RPC tools/call request
5. Printed the result to stdout

The AI agent does exactly the same thing — it just runs these commands via Bash.

Next Steps

• AI Agent Setup — integrate with Claude Code or Cursor
• Configuration — two-level config, server options
• Daemon Mode — keep heavy servers warm