#Custom Tools

Extend SciLaxy's capabilities by building custom tools. Tools can be added as builtin Python functions or provided externally via MCP servers.

#Tool Architecture

Tools in SciLaxy follow the LangChain tool interface. Each tool has:

• Name — Unique identifier used by the LLM to call the tool
• Description — Explains what the tool does (used by the LLM to decide when to use it)
• Schema — Input parameters with types and descriptions
• Implementation — The function that executes when the tool is called

#Building a Builtin Tool

Builtin tools are Python functions in service/app/tools/. Here's the structure:

# service/app/tools/my_tool/tools.py
from langchain_core.tools import tool

@tool
def my_custom_tool(query: str) -> str:
    """Search for information about a topic.

    Args:
        query: The search query to look up.
    """
    # Implementation here
    result = do_search(query)
    return result

#Tool with Schemas

For complex input parameters, define a Pydantic schema:

# service/app/tools/my_tool/schemas.py
from pydantic import BaseModel, Field

class MyToolInput(BaseModel):
    query: str = Field(description="The search query")
    max_results: int = Field(default=10, description="Maximum results to return")

# service/app/tools/my_tool/tools.py
from langchain_core.tools import tool
from .schemas import MyToolInput

@tool(args_schema=MyToolInput)
def my_custom_tool(query: str, max_results: int = 10) -> str:
    """Search for information with configurable result count."""
    results = do_search(query, limit=max_results)
    return format_results(results)

#Tool Registration

Tools are registered in the tool preparation step (service/app/tools/prepare.py). Add your tool to the tool list so agents can discover and use it:

from app.tools.my_tool.tools import my_custom_tool

# Add to the tool registry
BUILTIN_TOOLS = {
    "my_custom_tool": {
        "tool": my_custom_tool,
        "category": "search",
        "display_name": "My Custom Tool",
        "description": "Search for information about a topic",
    },
}

After registration, the tool appears in the agent configuration UI and can be enabled per-agent.

#Dynamic Tools via MCP

For tools that don't need to be part of the SciLaxy codebase, use MCP servers. This is the recommended approach for:

• Third-party API integrations
• User-specific tools
• Tools that change frequently
• Tools developed by external teams

See MCP Servers for details on connecting external tool providers.

#When to Use Builtin vs. MCP