js-mcp-server

js-mcp-server

2

The JS MCP Server is a component for easy deployment and integration of MCP protocols in JavaScript environments, utilizing WASM for efficient performance. It allows customization through custom handlers and supports interaction with Postgres databases and HTTP APIs using developer-friendly interfaces.

JS MCP Server

JavaScript MCP Server Component with WASM SDK for easy integration and deployment.

Setup

To get started with the JS MCP Server, follow these steps:

git clone https://github.com/justin-echternach/js-mcp-server.git
cd js-mcp-server
npm install -g @bytecodealliance/componentize-js @bytecodealliance/jco
npm install

Build

To build the project and generate the WebAssembly module:

npm run build

This will bundle the JavaScript code using Vite and componentize it into dist/js-mcp-server.wasm.

Customization

You can customize the server by adding or modifying handlers (also referred to as tools). Handlers are registered in the server.js file and are located in the handlers directory. To add a new handler:

  1. Create a new JavaScript file in the handlers directory with your handler implementation.
  2. Import your handler in server.js.
  3. Add your handler to the tools object in the createMcpServer function call, similar to how EchoTool, KgTool, and FetchDogTool are added.

For example, to add a new handler named MyCustomTool:

// In server.js
import { MyCustomTool } from './handlers/index.js';

createMcpServer({
  serverInfo: {
    name: "JavaScript MCP Server",
    version: "1.0.0"
  },
  tools: {
    [EchoTool.name]: EchoTool,
    [KgTool.name]: KgTool,
    [MyCustomTool.name]: MyCustomTool
  },
  handlers: {
    // Custom handlers (optional)
  }
});

Writing Custom Handlers

Handlers (tools) are the core of your server's functionality. Each handler should export:

  • An async handler function that processes requests and returns results.
  • A name (string), description, and schema for input validation and documentation.

Minimal Example

Here's a minimal handler based on FetchDogTool:

// handlers/fetch_dog_tool.js
export async function handleTool(parameters) {
  try {
    // Perform your logic (e.g., HTTP request)
    return { result: 'success!' };
  } catch (error) {
    return { error: error.message };
  }
}

export function getToolName() {
  return 'fetch_dog';
}

export function getToolDescription() {
  return 'Fetches a random dog image from the Dog API';
}

export function getToolSchema() {
  return { type: 'object', properties: {} };
}

export const FetchDogTool = {
  handler: handleTool,
  name: getToolName(),
  description: getToolDescription(),
  schema: getToolSchema()
};

Registering Your Handler

  1. Export it in handlers/index.js:
    import { FetchDogTool } from './fetch_dog_tool.js';
    export { FetchDogTool };
    // ...add to tools map if needed
    
  2. Add it to the tools object in server.js:
    import { FetchDogTool } from './handlers/fetch_dog_tool.js';
    createMcpServer({
      tools: {
        [FetchDogTool.name]: FetchDogTool,
        // ...other tools
      }
    });
    

Best Practices

  • Handlers can be async and return plain objects.
  • Always handle errors gracefully and return a clear error object.
  • Use the schema property to describe expected input.
  • Add meaningful name and description fields for discoverability.

For more advanced features (timeouts, retries, richer error handling), consider extending your handler logic accordingly.


Using the Postgres and HTTP Interfaces

Your MCP server can interact with external services such as HTTP APIs and Postgres databases using robust, developer-friendly abstractions.

Postgres Interface (Async)

You can query a Postgres database using the query function from the comapi:postgres/query@0.1.0 interface. This is useful for knowledge graph or data-backed tools.

Example:

import { query } from "comapi:postgres/query@0.1.0";

export async function handleTool(parameters) {
  const entityFilter = parameters?.entityFilter || "";
  const sqlQuery = `
    SELECT id, name FROM my_table
    WHERE name ILIKE $1
    LIMIT 10
  `;
  const inputValue = `%${entityFilter}%`;
  // PgValue enum/variant: { tag: 'text', val: value }
  const params = [{ tag: 'text', val: inputValue }];
  const result = await query(sqlQuery, params);
  return { rows: result.rows };
}

Best Practices:

  • WIT defined enums/variants must be wrapped as { tag: 'text', val: value } for text inputs.
  • Always validate and sanitize your inputs.
  • Handle errors with try/catch as needed.

HTTP Client Functions

The SDK provides a developer-friendly HTTP client abstraction for making outgoing HTTP requests, wrapping WASI HTTP complexity.

Example:

import { get } from '../sdk/http/index.js';

export async function handleTool(parameters) {
  try {
    // Simple GET request with JSON response
    const response = await get('https://dog.ceo/api/breeds/image/random', { json: true });
    return { image_url: response.message, status: response.status };
  } catch (error) {
    return { error: error.message };
  }
}

Available HTTP methods:

  • get(url, options)
  • post(url, data, options)
  • put(url, data, options)
  • patch(url, data, options)
  • del(url, options) (alias: deleteRequest)
  • head, options, postJson, putJson, patchJson, postForm, putForm
  • You can also create a custom HTTP client via createHttpClient.

Best Practices:

  • Use the json: true option to automatically parse JSON responses.
  • Handle errors with try/catch and return a clear error object.