haive.core.engine.aug_llm.mcp_config

MCP-enhanced AugLLMConfig with full type checking.

This module provides MCPAugLLMConfig, which extends AugLLMConfig with Model Context Protocol (MCP) support through proper mixin composition. It includes full type checking and seamless integration with the existing Haive configuration system.

The configuration automatically discovers MCP tools, manages resources, and enhances prompts while maintaining compatibility with all existing AugLLMConfig features.

Classes

MCPAugLLMConfig

AugLLMConfig enhanced with MCP (Model Context Protocol) support.

Functions

create_mcp_aug_llm_config(name[, model, mcp_servers])

Factory function to create and initialize MCPAugLLMConfig.

Module Contents

class haive.core.engine.aug_llm.mcp_config.MCPAugLLMConfig(/, **data)

Bases: _get_mcp_mixin(), haive.core.engine.aug_llm.config.AugLLMConfig

AugLLMConfig enhanced with MCP (Model Context Protocol) support.

This configuration class extends AugLLMConfig with MCP capabilities through the MCPMixin, providing: - Automatic MCP tool discovery and integration - Resource management from MCP servers - Prompt template enhancement - Full type safety with MCP configurations

The class properly integrates with ToolRouteMixin (inherited from AugLLMConfig) to ensure MCP tools are correctly routed and managed alongside regular tools.

Parameters:

data (Any)

All attributes from AugLLMConfig plus

- mcp_config

Optional MCP configuration

- mcp_resources

List of discovered MCP resources

- mcp_prompts

Dictionary of MCP prompt templates

- auto_discover_mcp_tools

Whether to auto-discover tools

- inject_mcp_resources

Whether to inject resources

- use_mcp_prompts

Whether to use MCP prompts

Example

Basic usage with MCP server:

from haive.mcp.config import MCPConfig, MCPServerConfig

config = MCPAugLLMConfig(
    name="mcp_agent",
    llm_config=LLMConfig(provider="openai", model="gpt-4"),
    mcp_config=MCPConfig(
        enabled=True,
        servers={
            "filesystem": MCPServerConfig(
                transport="stdio",
                command="npx",
                args=["-y", "@modelcontextprotocol/server-filesystem"]
            )
        }
    ),
    system_message="AI assistant with filesystem access"
)

# Initialize MCP
await config.setup()

# MCP tools are now available alongside regular tools
print(f"Total tools: {len(config.tools)}")
print(f"MCP tools: {len(config.get_mcp_tools())}")

Multiple MCP servers:

config = MCPAugLLMConfig(
    name="multi_mcp_agent",
    llm_config={"provider": "openai", "model": "gpt-4"},
    mcp_config=MCPConfig(
        enabled=True,
        servers={
            "filesystem": MCPServerConfig(...),
            "github": MCPServerConfig(...),
            "postgres": MCPServerConfig(...)
        }
    ),
    tools=["calculator"],  # Regular tools work too
    auto_discover_mcp_tools=True
)

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

cleanup()

Clean up resources including MCP connections.

Should be called when the configuration is no longer needed.

Return type:

None

debug_mcp_state()

Print debug information about MCP integration state.

Useful for troubleshooting MCP configuration and tool discovery.

Return type:

None

get_all_tools()

Get all available tool names including MCP tools.

Returns:

Combined list of regular tools and MCP tool names

Return type:

list[str]

get_tool_by_name(name)

Get a tool instance by name, checking both regular and MCP tools.

Parameters:

name (str) – Tool name to retrieve

Returns:

Tool instance or None if not found

Return type:

Any | None

async setup()

Initialize both AugLLMConfig and MCP integration.

This method: 1. Sets up MCP servers and discovers tools/resources/prompts 2. Enhances the system message with MCP information 3. Integrates MCP tools with the existing tool routing system

Should be called after creating the configuration but before using it with an agent.

Return type:

None

async haive.core.engine.aug_llm.mcp_config.create_mcp_aug_llm_config(name, model='gpt-4', mcp_servers=None, **kwargs)

Factory function to create and initialize MCPAugLLMConfig.

Parameters:

• name (str) – Configuration name

• model (str) – LLM model to use

• mcp_servers (dict[str, Any] | None) – Dictionary of MCP server configurations

• **kwargs – Additional arguments for MCPAugLLMConfig

Returns:

Initialized MCPAugLLMConfig instance

Return type:

MCPAugLLMConfig

Example

Creating a configuration with factory:

config = await create_mcp_aug_llm_config(
    name="assistant",
    model="gpt-4",
    mcp_servers={
        "filesystem": {
            "transport": "stdio",
            "command": "npx",
            "args": ["-y", "@modelcontextprotocol/server-filesystem"]
        }
    },
    system_message="Helpful AI assistant",
    temperature=0.7
)