Feature/mcp prompts support #183

brodguez-atlas · 2025-06-25T16:21:44Z

Add MCP Prompts Support to FastAPI-MCP

Summary

This PR adds comprehensive MCP (Model Context Protocol) prompts support to FastAPI-MCP, allowing users to create custom prompts and auto-generate helpful prompts for their FastAPI endpoints.

Features Added

Auto-generated tool prompts: Automatically creates prompts for each FastAPI endpoint
Custom prompt support: Users can define custom prompts with the @mcp.prompt() decorator
Prompt override capability: Custom prompts can override auto-generated ones
Flexible prompt control: Users can enable/disable auto-generation as needed

Changes Made

Added fastapi_mcp/prompts.py with core prompt functionality
Enhanced FastApiMCP class to support prompt registration and management
Added comprehensive tests in tests/test_prompts.py
Updated documentation and examples
Added support for both sync and async prompt functions

Testing

All existing tests pass
Added 16+ new tests covering prompt functionality
Coverage maintained at 84.60%
Linting and type checking passes

Usage Example

from fastapi import FastAPI
from fastapi_mcp import FastApiMCP

app = FastAPI()
mcp = FastApiMCP(app, auto_generate_prompts=True)

@mcp.prompt("api_help")
def api_help_prompt():
    return PromptMessage(
        role="user",
        content=TextContent(text="How can I use this API effectively?")
    )

mcp.mount()

- Add PromptRegistry class for managing prompt functions - Add @mcp.prompt() decorator for FastAPI-native prompt registration - Add prompt-related types (PromptMessage, TextContent, ImageContent, etc.) - Add comprehensive test suite for prompts functionality - Add example demonstrating various prompt patterns - Support both sync and async prompt functions - Support parameterized prompts with type validation - Integrate with FastAPI dependency injection system - Follow MCP specification for prompts/list and prompts/get

- Introduced detailed documentation for auto-generated and custom prompts in README.md. - Enhanced examples to demonstrate the usage of MCP prompts, including tool prompts and troubleshooting prompts. - Implemented auto-registration of tool prompts in the FastApiMCP server for improved user guidance. - Added tests to verify the functionality of auto-generated prompts.

- Updated prompt titles and descriptions to enhance user understanding. - Revised the structure of auto-generated prompts to focus on actionable guidelines and best practices. - Adjusted tests to reflect changes in prompt content and ensure accuracy in generated messages.

- Consolidated prompt parameters in examples for cleaner presentation. - Enhanced spacing and structure in the PromptRegistry class for better clarity. - Updated test cases to reflect formatting changes and ensure consistency.

- Updated README.md to clarify the default behavior of auto-generated prompts and added examples for controlling prompt generation. - Improved the example script to demonstrate three approaches for managing prompts: auto-generated only, custom prompts only, and a mixed approach. - Modified the FastApiMCP class to include an option for enabling or disabling auto-prompt generation. - Added tests to verify the functionality of prompt auto-generation control, ensuring both custom and auto-generated prompts work as intended.

boris-42

This looks very promissing. Thank you for your contirubtion: @brodguez-atlas

Note I am not a maintiner. Just trying to help with reviews.

I would suggest for next PRs to avoid unnecessary style changes. If you make them as separate PR they will be merged same day (hopefully) and your main PR less scary to review.

I pointed few things please take a look

boris-42 · 2025-07-15T23:38:22Z

examples/03_custom_exposed_endpoints_example.py

@@ -6,7 +6,8 @@
 - You can combine operation filtering with tag filtering (e.g., use `include_operations` with `include_tags`)
 - When combining filters, a greedy approach will be taken. Endpoints matching either criteria will be included
 """
-from examples.shared.apps.items import app # The FastAPI app
+


@brodguez-atlas It's better to avoid unnecessary style changes with features as it makes it harder to review. 2 separated PR would be much better.

boris-42 · 2025-07-15T23:41:20Z

examples/10_prompts_example.py

+    title="Prompts Example API", description="An example API demonstrating MCP Prompts functionality", version="1.0.0"
+)
+
+# Create MCP server with auto-generated prompts enabled (default: True)


It should be defaulted to the False, to preserve original behavior, and not impact existing integrations.

boris-42 · 2025-07-15T23:42:53Z

examples/10_prompts_example.py

+    """Generate a welcome message for API users."""
+    return PromptMessage(
+        role="user",
+        content=TextContent(text="Please provide a warm and friendly welcome message for new users of our API."),


Can you elaborate how LLM will consume this?

boris-42 · 2025-07-15T23:56:40Z

fastapi_mcp/server.py

+        auto_generate_prompts: Annotated[
+            bool,
+            Doc("Whether to automatically generate default prompts for each tool. Defaults to True."),
+        ] = True,


Make it False to preserve existing behavior.

boris-42 · 2025-07-16T00:05:19Z

fastapi_mcp/prompts.py

+        """
+        # Focus on actionable guidance rather than repeating tool information
+        content_parts = [
+            f"You are about to use the **{tool.name}** tool.",


can this be done once for all tools? it looks more like MCP server description, rather than tool description..

boris-42 · 2025-07-16T00:09:25Z

fastapi_mcp/prompts.py

+                        return self._type_to_json_schema(non_none_type)
+
+        # Default to string for unknown types
+        return {"type": "string"}


I would raise here errors, this looks like a ticking bomb to me

boris-42 · 2025-07-16T00:10:43Z

fastapi_mcp/prompts.py

+                arguments.append(PromptArgument(name=param_name, description=param_desc, required=is_required))
+
+                # Create JSON schema property
+                properties[param_name] = self._type_to_json_schema(param_type)


let's just use pydantic here, if we pass data: (somepaynanticclass) it will automatically have funcitonallity to generate json schema and it will be well aligned to the fastapi and less magic

brodguez-atlas added 6 commits June 25, 2025 14:52

Refactor prompt formatting in examples for improved readability

fb8ae82

Refactor prompt formatting and improve code readability

e760ea1

- Consolidated prompt parameters in examples for cleaner presentation. - Enhanced spacing and structure in the PromptRegistry class for better clarity. - Updated test cases to reflect formatting changes and ensure consistency.

boris-42 reviewed Jul 16, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Feature/mcp prompts support #183

Feature/mcp prompts support #183

Uh oh!

brodguez-atlas commented Jun 25, 2025

Uh oh!

boris-42 left a comment

Uh oh!

boris-42 Jul 15, 2025

Uh oh!

boris-42 Jul 15, 2025

Uh oh!

boris-42 Jul 15, 2025

Uh oh!

boris-42 Jul 15, 2025

Uh oh!

boris-42 Jul 16, 2025

Uh oh!

boris-42 Jul 16, 2025

Uh oh!

boris-42 Jul 16, 2025

Uh oh!

Uh oh!

Feature/mcp prompts support #183

Are you sure you want to change the base?

Feature/mcp prompts support #183

Uh oh!

Conversation

brodguez-atlas commented Jun 25, 2025

Add MCP Prompts Support to FastAPI-MCP

Summary

Features Added

Changes Made

Testing

Usage Example

Uh oh!

boris-42 left a comment

Choose a reason for hiding this comment

Uh oh!

boris-42 Jul 15, 2025

Choose a reason for hiding this comment

Uh oh!

boris-42 Jul 15, 2025

Choose a reason for hiding this comment

Uh oh!

boris-42 Jul 15, 2025

Choose a reason for hiding this comment

Uh oh!

boris-42 Jul 15, 2025

Choose a reason for hiding this comment

Uh oh!

boris-42 Jul 16, 2025

Choose a reason for hiding this comment

Uh oh!

boris-42 Jul 16, 2025

Choose a reason for hiding this comment

Uh oh!

boris-42 Jul 16, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!