MCP-Less / Code-mode

Talk to ANY MCP directly (no client integration required)

You can call MCP servers directly from CLI/code with bunx — no Pi/Claude/Cursor connector setup required. Or write scripts/code for this flow.

Method 1: Direct calls with MCP Inspector CLI

Use @modelcontextprotocol/inspector in CLI mode.

A) Remote Streamable HTTP MCP

# List tools
bunx --yes @modelcontextprotocol/inspector \
  --cli https://your-mcp-server.example.com/mcp \
  --transport http \
  --method tools/list

# Call tool (no args)
bunx --yes @modelcontextprotocol/inspector \
  --cli https://your-mcp-server.example.com/mcp \
  --transport http \
  --method tools/call \
  --tool-name health_check

# Call tool (with args)
bunx --yes @modelcontextprotocol/inspector \
  --cli https://your-mcp-server.example.com/mcp \
  --transport http \
  --method tools/call \
  --tool-name my_tool \
  --tool-arg 'query=hello world' \
  --tool-arg 'limit=10'

B) Remote SSE MCP

bunx --yes @modelcontextprotocol/inspector \
  --cli https://your-mcp-server.example.com/sse \
  --transport sse \
  --method tools/list

C) Local stdio MCP process

# Example: node server
bunx --yes @modelcontextprotocol/inspector \
  --cli node ./dist/index.js \
  --method tools/list

Method 2: Bridge remote MCP to stdio clients (optional)

If a client only supports stdio servers, proxy a remote MCP via mcp-remote:

bunx --yes mcp-remote https://your-mcp-server.example.com/mcp --transport http-first

This runs a local stdio proxy connected to the remote MCP server.

Auth headers (when needed)

For token-protected HTTP/SSE servers:

bunx --yes @modelcontextprotocol/inspector \
  --cli https://your-mcp-server.example.com/mcp \
  --transport http \
  --header "Authorization: Bearer $TOKEN" \
  --method tools/list

Practical workflow for agents

1. tools/list
2. Optional capability/discovery tool (often read_me, help, or describe)
3. tools/call with minimal valid args
4. Iterate using server errors/schema hints