MCP Tool Creator

This skill guides you through creating new MCP tools for the Datadog MCP server with proper typing, documentation, and architecture.

Step-by-Step Tool Creation

Step 1: Plan the Tool

Before writing code, determine:

•Purpose: What specific task does this tool accomplish?
•API Endpoint: Which Datadog API endpoint(s) will it use?
•Parameters: What inputs does it need?
•Return Type: What data structure will it return?
•Domain: Which tool file should contain it? (logs, metrics, dashboards, apm, monitors, or new domain)

Step 2: Define TypedDict Response Structure

Create explicit types before implementation:

python

from typing import TypedDict, NotRequired

class ToolResponse(TypedDict):
    """Response from the new tool."""
    success: bool
    error: NotRequired[str]
    # Add specific fields for your tool
    data: list[DataItem]
    count: int
    next_cursor: NotRequired[str | None]

Step 3: Create Internal Implementation

In the appropriate tools/*.py file:

python

"""Module docstring describing this domain."""

from typing import TypedDict, Literal, NotRequired
from datadog_api_client.v2.api.relevant_api import RelevantApi
from datadog_api_client.v2.model.request_model import RequestModel

from ..auth import DatadogAuth
from ..utils.response import ResponseBuilder, format_error_response
from ..utils.auth import get_api_instance

# Define response types first
class ItemData(TypedDict):
    """Individual item in response."""
    id: str
    name: str
    value: int

class ToolResponse(TypedDict):
    """Complete tool response."""
    success: bool
    data: list[ItemData]
    count: int
    error: NotRequired[str]

# Implementation function
def my_new_tool(
    param1: str,
    param2: int | None = None,
    auth: DatadogAuth | None = None
) -> ToolResponse:
    """Internal implementation with business logic.

    Args:
        param1: Description of parameter 1
        param2: Description of parameter 2 (optional)
        auth: DatadogAuth instance (injected)

    Returns:
        ToolResponse with data and metadata
    """
    # Get API instance
    api_instance, auth = get_api_instance(RelevantApi, auth)

    try:
        # Build API request
        request = RequestModel(
            param1=param1,
            param2=param2
        )

        # Call API
        response = api_instance.call_endpoint(body=request)

        # Format data
        items: list[ItemData] = []
        if hasattr(response, 'data') and response.data:
            for item in response.data:
                items.append({
                    "id": item.id if hasattr(item, 'id') else "",
                    "name": item.name if hasattr(item, 'name') else "",
                    "value": item.value if hasattr(item, 'value') else 0
                })

        # Use ResponseBuilder for auto-truncation
        return ResponseBuilder.success("data", items)

    except Exception as e:
        return format_error_response("data", e)

Step 4: Add FastMCP Tool Decorator in server.py

python

# In server.py
from datadog_mcp.tools.domain import my_new_tool as _my_new_tool

@mcp.tool()
def my_new_tool(
    param1: str,
    param2: int | None = None
) -> dict[str, object]:
    """One-line summary of what this tool does.

    ⚠️ IMPORTANT USAGE NOTES: When to use this tool vs alternatives.

    Use this when:
    - Specific use case 1
    - Specific use case 2

    DO NOT use for:
    - Alternative tool 1 (use other_tool instead)
    - Alternative tool 2 (use another_tool instead)

    Args:
        param1: Clear description with examples (e.g., "user_id" or "email@example.com")
        param2: Clear description with range/constraints (default: None, range: 1-100)

    Returns:
        dict with success flag, data array, and count

    Examples:
        my_new_tool("value1")
        my_new_tool("value1", param2=50)
    """
    auth = get_auth_instance()
    return _my_new_tool(param1, param2, auth=auth)

Step 5: Validation Checklist

Before committing, verify:

Typing:

• No Any types
• All dict and list have type parameters
• TypedDict defined for response structure
• Function signature has complete type hints
• Using Python 3.10+ syntax (no List, Dict, Optional)

Documentation:

• Tool has comprehensive docstring
• Includes "Use this when" section
• Includes "DO NOT use for" section
• All parameters documented with examples
• Return value documented
• Usage examples provided

Architecture:

• Implementation in tools/*.py file
• FastMCP decorator in server.py
• Uses ResponseBuilder.success()
• Uses format_error_response() for errors
• Follows dependency injection pattern

Safety:

• Tool is read-focused (GET-equivalent)
• Destructive operations require confirmation
• Input validation implemented
• Error handling comprehensive

Tool Naming Guidelines

Choose clear, consistent names:

python

# ✅ GOOD
search_logs()
query_metrics()
list_dashboards()
create_dashboard()
delete_monitor()
count_logs()

# ❌ BAD
searchDatadogLogsWithQuerySyntax()  # Too verbose
search()  # Too vague
get_stuff()  # Unclear

Pattern: verb_noun (e.g., search_logs, create_dashboard, count_metrics)

Common Patterns

Pattern 1: Search/List Tool

python

def search_items(
    query: str,
    page_size: int = 25,
    cursor: str | None = None,
    auth: DatadogAuth | None = None
) -> dict[str, object]:
    """Search with pagination."""
    from ..utils.pagination import validate_page_size

    page_size = validate_page_size(page_size)
    api_instance, auth = get_api_instance(ItemsApi, auth)

    try:
        response = api_instance.search(query=query, limit=page_size, cursor=cursor)

        items = [format_item(item) for item in response.data]

        return ResponseBuilder.success(
            "items",
            items,
            next_cursor=response.next_cursor if hasattr(response, 'next_cursor') else None,
            has_more=response.has_more if hasattr(response, 'has_more') else False
        )
    except Exception as e:
        return format_error_response("items", e)

Pattern 2: Aggregation/Count Tool

python

def count_items(
    query: str,
    from_time: str,
    to_time: str,
    auth: DatadogAuth | None = None
) -> dict[str, object]:
    """Count without fetching data."""
    api_instance, auth = get_api_instance(ItemsApi, auth)

    try:
        response = api_instance.aggregate(
            query=query,
            from_time=from_time,
            to_time=to_time,
            aggregation="count"
        )

        count = extract_count(response)

        return {
            "success": True,
            "count": count,
            "query": query
        }
    except Exception as e:
        return {
            "success": False,
            "error": str(e),
            "count": 0
        }

Pattern 3: Create/Update Tool

python

def create_item(
    name: str,
    config: dict[str, str | int | bool],
    tags: list[str] | None = None,
    auth: DatadogAuth | None = None
) -> dict[str, object]:
    """Create new item."""
    api_instance, auth = get_api_instance(ItemsApi, auth)

    try:
        request = CreateItemRequest(
            name=name,
            config=config,
            tags=tags or []
        )

        response = api_instance.create(body=request)

        return {
            "success": True,
            "item_id": response.id,
            "url": response.url if hasattr(response, 'url') else None
        }
    except Exception as e:
        return {
            "success": False,
            "error": str(e)
        }

Integration Testing

After creating the tool, test it:

python

# Manual test
from datadog_mcp.tools.domain import my_new_tool
from datadog_mcp.auth import DatadogAuth

auth = DatadogAuth()
result = my_new_tool("test_param", auth=auth)
print(result)

Documentation Updates

Update the README.md with the new tool:

markdown

### Domain Name
- **my_new_tool** - Brief description of what it does and when to use it

Summary

Creating a new tool requires:

•Define TypedDict response structure
•Implement in tools/*.py with full typing
•Add FastMCP decorator in server.py with rich docs
•Use ResponseBuilder and format_error_response
•Follow dependency injection pattern
•Test manually
•Update README.md

This ensures consistency, type safety, and excellent developer experience.