Summoning heroes and loading quests...
An MCP (Model Context Protocol) server is a lightweight service that gives AI agents standardised access to external tools, data sources, and APIs. It acts as the universal bridge between a language model and the real-world systems it needs to read, write, and control.
Video tutorial — coming soon
What is an MCP Server?
When an AI agent needs to call an external tool, it sends a structured request to the MCP server, which routes it to the correct API and returns a typed response. This keeps the model's context clean and the tool integrations reusable across any agent.
MCP servers implement a standardised JSON-RPC protocol so any compliant AI runtime can connect without custom glue code. Each tool is described with a schema the model reads at runtime, enabling dynamic tool discovery without prompt engineering.
Remote MCP servers are managed services you connect to instantly via a URL, ideal for rapid prototyping and public tools. Self-hosted servers run inside your own infrastructure, giving you full data control, private network access, and no usage limits.
For most operators, a remote server is the fastest path to a working agent. Developers with compliance requirements, proprietary data, or custom tool needs should self-host using the Binary Heros MCP toolkit, deployable via Docker in under five minutes.
Traditional APIs require hand-coded integration per tool and per model, with no standard for how a model discovers or calls them. MCP standardises the interface so one server implementation works with any MCP-compatible AI runtime — Claude, GPT-4o, Gemini, and open-source models alike.
The key difference is runtime discoverability. A model can query an MCP server to learn what tools are available and call them dynamically based on the task. With plain REST APIs, every tool must be described in the system prompt before the model can use it.
Follow these steps in order to get started.
Install the MCP runtime
Run `npx @binary-heros/mcp init` to scaffold your server folder with TypeScript config and example tools.
Define your tools
Describe each tool with a name, JSON input schema, and async handler function in `tools/index.ts`.
Connect a model
Point any MCP-compatible AI client at `http://localhost:3100` or your deployed URL to start calling tools.
Test with the inspector
Use the built-in MCP inspector at `/inspect` to fire test calls, view typed responses, and debug errors.
Deploy and share
Push to Railway or Fly.io in one command. Paste the server URL into the Binary Heros MCP directory to share it.
Mission Template
Scaffold a working MCP server with two custom tools, connect it to an AI agent, and deploy it publicly. You will implement the core MCP primitives: Resources, Tools, and Prompts.