Package Exports

@meldscience/mcp-stream-parser
@meldscience/mcp-stream-parser/dist/index.js

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (@meldscience/mcp-stream-parser) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

MCP Stream Parser

Extends the Anthropic SDK to provide structured parsing of Claude's responses. It processes Claude's streaming responses into clearly delineated types of messages while maintaining full compatibility with the SDK's types and events.

Features

Structured Message Parsing
- Regular messages (plain text)
- Thought messages (Claude's reasoning)
- Tool messages (function calls)
- Image content handling
Dual Operating Modes
- Message Mode: Complete semantic units for platforms like Discord/Slack
- Streaming Mode: Real-time updates for interactive UIs
Type-Safe Event System
- Full SDK type compatibility
- Strongly-typed events
- Error recovery support

Installation

npm install @meldscience/mcp-stream-parser

Quick Start

import { MCPStreamParser } from '@meldscience/mcp-stream-parser';
import { Client } from '@anthropic-ai/sdk';

// Initialize the parser
const parser = new MCPStreamParser();

// Handle different message types
parser.on('message', ({ content }) => {
  console.log('Message:', content);
});

parser.on('thought', ({ content, hasToolCall }) => {
  console.log('Thought:', content);
});

parser.on('tool', ({ name, params, thoughtContent }) => {
  console.log('Tool:', name, params);
});

// Process a Claude stream
const client = new Client(process.env.ANTHROPIC_API_KEY);
const response = await client.messages.create({
  model: 'claude-3-opus-20240229',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello!' }],
  stream: true
});

await parser.processStream(response.stream);

Configuration

const parser = new MCPStreamParser({
  // Parser mode
  mode: 'message', // or 'stream'

  // Message handling (for platforms like Discord)
  message: {
    maxLength: 2000,
    overflow: 'split',
    splitPoints: ['\n\n', '. ', ' ']
  },

  // Buffer safety
  buffer: {
    maxSize: 64 * 1024,
    overflow: 'error'
  }
});

Documentation

User Guide - Detailed usage and examples
Implementation - Code structure and data flows
Development Plan - Implementation strategy

Error Handling

parser.on('error', ({ error, fatal, partialContent }) => {
  if (fatal) {
    console.error('Fatal error:', error);
    parser.dispose();
  } else {
    console.warn('Recoverable error:', error);
    if (partialContent) {
      console.log('Partial content:', partialContent);
    }
  }
});

Features

Message Types

Regular Messages: Plain text content from Claude
Thought Messages: Claude's internal reasoning
Tool Messages: Function calls and tool usage
Image Content: Pass-through handling

Operating Modes

Message Mode: Complete semantic units
Streaming Mode: Real-time character updates

Event System

Text events (streaming)
Message events (complete units)
Thought events (reasoning)
Tool events (function calls)
Error events (with recovery)

License

MIT License - See LICENSE file for details