Haive MCP Documentation¶

Welcome to Haive MCP¶

Haive MCP (Model Context Protocol) revolutionizes how AI agents acquire capabilities by enabling dynamic, runtime integration of tools, resources, and prompts from a vast ecosystem of 1900+ MCP servers from top GitHub repositories.

🚀 The Game Changer: Your agents can now automatically discover and integrate the exact tools they need, when they need them, without any predefined configuration!

Dynamic Integration at Its Core¶

Imagine an agent that starts with zero tools but can:

  • Self-query through 992+ servers to find capabilities

  • Automatically install and configure needed tools at runtime

  • Dynamically integrate tools, resources, and prompts into its workflow

  • Adapt capabilities based on the task at hand

Example: Ask your agent to “analyze this GitHub repo and create visualizations” - it will automatically discover and integrate GitHub tools, code analysis tools, and visualization libraries without any manual setup!

Key Features¶

  • 🔍 Self-Query Retrieval - Agents search through 1900+ servers using intelligent matching

  • ⚡ Dynamic Tool Integration - Tools are discovered, installed, and integrated at runtime

  • 🧠 AI-Powered Discovery - Smart recommendations based on task analysis

  • 🔌 Zero Configuration - Many servers work instantly via npx

  • 🎯 Capability Matching - Find servers by what they can do, not by name

  • 🔄 Runtime Adaptation - Agents evolve their capabilities during execution

  • đź“š Beyond Tools - Also discovers resources (databases, APIs) and domain-specific prompts

  • 🛡️ HITL Workflows - Optional human approval for security

  • đź’Ş Automatic Failover - Finds alternative servers if primaries fail

What is MCP?¶

Model Context Protocol (MCP) is an open standard that allows AI systems to securely connect to data sources and tools. With MCP, your agents can:

  • Access file systems and databases

  • Use web search and APIs

  • Control development tools

  • Interact with specialized services

  • And much more!

Quick Example: Dynamic Discovery in Action¶

Watch an agent automatically discover and integrate tools¶
from haive.mcp.agents import IntelligentMCPAgent
from haive.core.engine import AugLLMConfig

# Create an intelligent agent with dynamic discovery
agent = IntelligentMCPAgent(
    engine=AugLLMConfig(),
    auto_discover=True  # Enable automatic tool discovery
)

# Ask it to do something - it will find the tools it needs!
result = await agent.arun({
    "messages": [{
        "role": "user",
        "content": "Search GitHub for Python AI projects and create a report"
    }]
})

Note

The agent automatically:

  1. Discovers GitHub access → finds @modelcontextprotocol/server-github

  2. Discovers file writing → finds @modelcontextprotocol/server-filesystem

  3. Finds code analysis → locates relevant analysis servers

  4. Installs all tools dynamically at runtime

  5. Completes the task with newly acquired capabilities!

Massive Server Discovery Database¶

🔍 Smart Discovery

Ask for any capability and agents find the right tools:

  • “Extract data from PDFs” → PDF processing servers

  • “Connect to PostgreSQL” → Database servers

  • “Create visualizations” → Charting tools

  • “Send emails” → Email API servers

  • “Scrape websites” → Web scraping tools

🚀 Vast Ecosystem

Access to 1900+ servers from top repositories:

  • Financial data & crypto analysis

  • Smart home & IoT integration

  • Medical image processing

  • Document translation

  • Cloud resource optimization

Getting Help¶

Next Steps¶

Indices and tables¶