mcpField
M
mcp-connector

mcp-connector

github.com/starchild-ai-agentAgent workflows3094 installsUpdated 1w ago
Repository

About mcp-connector

A skill for connecting to and interacting with MCP servers. It supports registry-based discovery as well as ad-hoc connections, then exposes flows to inspect tools and resources, call tools, and manage multiple session connections.

Key features

  1. 01Browse a preconfigured MCP registry
  2. 02Support stdio, SSE, and Streamable HTTP
  3. 03Discover tools and resources
  4. 04Call tools and read resources
  5. 05Connect and disconnect multiple servers

Use cases

  1. 01Find MCP servers by category
  2. 02Connect to servers outside the registry
  3. 03Use tools from multiple servers in one session

MCP support

The skill acts as an MCP client, connecting to local or remote MCP servers to call their tools and read resources.

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/starchild-ai-agent/official-skills --skill mcp-connector

After installing, type/mcp-connectorin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

__init__.py
client.py
manager.py
registry.py
tools.py

Details

From the project README

MCP Connector

Connect to any MCP (Model Context Protocol) server and interact with its tools, resources, and prompts.

Supported Transports

  • stdio — Launch a local MCP server as a subprocess (e.g. npx @modelcontextprotocol/server-filesystem /path)
  • sse — Connect to a remote MCP server via HTTP Server-Sent Events
  • streamable-http — Connect via the newer streamable HTTP transport (POST with SSE response)

Usage Flow

  1. Browse available servers with mcp_browse (or filter by category)
  2. Connect by name from the registry: mcp_connect(name="coingecko")
  3. Discover available tools/resources with mcp_list_tools or mcp_list_resources
  4. Use them via mcp_call_tool or mcp_read_resource
  5. Disconnect when done with mcp_disconnect

Pre-configured Registry

A curated list of MCP servers is maintained in config/mcp_servers.yaml. Connect to any of them by name — no URL or transport details needed:

mcp_connect(name="coingecko")     # Crypto market data
mcp_connect(name="exa")           # Neural web search
mcp_connect(name="dexpaprika")    # DEX data across chains

Use mcp_browse() to see all available servers, or filter by category:

mcp_browse(category="blockchain")
mcp_browse(category="developer")

Some servers require environment variables (API keys). The tool will tell you which ones are missing.

Ad-hoc Connections

You can still connect to any MCP server not in the registry by providing transport details:

stdio (local subprocess)
mcp_connect(
    name="fs",
    transport="stdio",
    command="npx",
    args=["-y", "@modelcontextprotocol/server-filesystem", "/workspace"]
)
SSE (HTTP Server-Sent Events)
mcp_connect(
    name="remote-tools",
    transport="sse",
    url="https://mcp.example.com/sse"
)
Streamable HTTP
mcp_connect(
    name="api-server",
    transport="streamable-http",
    url="https://mcp.example.com/mcp"
)

Using Connected Servers

Call a tool
mcp_call_tool(
    server="coingecko",
    tool="get_price",
    arguments={"coin": "bitcoin"}
)
Read a resource
mcp_read_resource(
    server="remote-tools",
    uri="file:///data/config.json"
)

Notes

  • Connections persist for the duration of the session
  • Multiple servers can be connected simultaneously
  • Server names must be unique — reconnecting with the same name replaces the old connection
  • Environment variables can be passed to stdio servers via the env parameter
  • Registry entries are in config/mcp_servers.yaml — add new servers there (no code changes needed)

Comments

Markdown supported