MCP Coding Agents

You run a transport-agnostic MCP server that coordinates models and AI agents, enabling real-time context management across STDIO and HTTP transports. It supports Claude Code, Gemini, and Codex integrations, built-in tools, and streaming capabilities for interactive workflows and editor integrations.

How to use

You can run this MCP server in STDIO mode for direct process communication, or in HTTP mode for network access. Use the two primary entry points to connect your client: a local STDIO process or a remote HTTP endpoint. In HTTP mode, the server exposes a JSON-RPC endpoint and can stream notifications via SSE, enabling real-time interactions with your AI agents.

How to install

Prerequisites: Node.js and npm installed on your system.

Install the MCP server globally via npm.

You can start a quick session using npx without installing the package locally.

Quick steps to get started in HTTP mode (default port 3050):

Step-by-step setup

npm install -g @kadreio/mcp-coding-agents

# Run in default HTTP mode on port 3050
npx @kadreio/mcp-claude-code

# Or run in STDIO mode for direct process communication
npx @kadreio/mcp-claude-code stdio

# Run in HTTP mode on a custom port (example: 3051)
npx @kadreio/mcp-claude-code http --port 3051

# Run in HTTP mode binding to a specific host (example: 0.0.0.0)
npx @kadreio/mcp-claude-code http --host 0.0.0.0

Configuration and runtime options

Configure the server to enable HTTPS, set the port, and manage session storage. You can also connect clients through the HTTP endpoint or via STDIO for local tooling.

HTTPS support

You can enable HTTPS in HTTP mode and supply your own certificates or use an auto-generated self-signed certificate for development.

MCP client connection methods

Two primary connection methods are supported: a remote HTTP endpoint and a local STDIO process. The HTTP method uses a URL to the MCP endpoint and supports streaming, while the STDIO method runs the MCP as a local process and communicates over standard input and output.

Notes on environment and usage

Environment variables control core configuration and agent behavior. You will typically set API keys for Claude Code, model choices, and file paths for executables. Ensure your environment is prepared for any required native dependencies and that the HTTP port you choose is accessible to your clients.

Summary

This MCP server combines transport flexibility with multi-agent capabilities and practical tools, making it suitable for local development, editor integrations, and networked deployments.

Configuration files and examples

The server can be configured via programmatic JSON structures for MCP clients in both STDIO and HTTP modes. These configurations specify how to connect to the server, which agent tools to enable, and any required environment variables.

Security and best practices

Use HTTPS in production to protect data in transit. Keep API keys secret and rotate them regularly. If running locally, prefer private networks for HTTP access and limit CORS as appropriate.

Available tools

calculate_bmi

Compute Body Mass Index from height and weight values.

get_timestamp

Return the current timestamp.

execute_command

Execute a shell command with optional timeout and capture output.

stream_sse_timestamps

Generate streaming timestamps via Server-Sent Events in HTTP mode.

claude_code_query

Query Claude Code with support for streaming responses.

gemini_query

Execute Gemini CLI commands.

codex_query

Execute Codex with JSONL streaming.