Skip to main content
Glama
AI-Project-Planner-Agent

AI-Project-Planner-Agent

by AdiRatnam

AI Project Feasibility & Solution Planner Agent

An MCP-based AI agent that evaluates an AI project idea using structured reasoning stages. The project focuses on reasoning quality, prompt engineering, structured outputs, and clean logs using the Gemini API (optimized for Free Tier).

Screenshots

Related MCP server: Gemini DeepSearch MCP

Architecture

The project consists of a FastMCP server and a client interacting with the Gemini 2.5 Flash model.

  • Client (client.py): Prompts the user for a project idea, communicates with the Gemini model, routes tool calls to the MCP server, and formats the output cleanly using the Rich library.

  • Server (server.py): A FastMCP server exposing a research_topic tool that uses DuckDuckGo Search to fetch up-to-date web information.

  • Models (models/reasoning_models.py): Pydantic models ensuring structured JSON outputs for all 7 reasoning stages.

  • Reasoning: Evaluates the idea through 7 structured stages:

    1. Problem Understanding

    2. Requirement Analysis

    3. Complexity & Constraint Analysis

    4. Research Decision (Calls MCP tool only if needed)

    5. Technology Recommendation

    6. Risk & Roadmap

    7. Final Report

Folder Structure

s5 - AI Project Planner Agent/
│
├── server.py               # FastMCP server exposing the research_topic tool
├── client.py               # Main client application handling Gemini interactions
├── prompts/
│   └── system_prompt.md    # System prompt guiding the reasoning stages
├── tools/
│   └── research_tool.py    # Implementation of web search logic
├── models/
│   └── reasoning_models.py # Pydantic models for structured outputs
├── README.md               # Project documentation
└── pyproject.toml          # uv configuration

Installation

This project uses uv for lightning-fast dependency management.

# Sync dependencies
uv sync

(If initializing from scratch):

uv init
uv add google-genai mcp fastmcp pydantic rich duckduckgo-search python-dotenv

Environment Variables: Create a .env file in the root directory and add your Gemini API key (ensure .env is listed in your .gitignore!):

GEMINI_API_KEY="your_api_key_here"

Usage

You do not need to run the server separately. The client automatically spawns the MCP server over stdio using uv.

Running the Client

uv run client.py

Example User Prompts

I want to build an AI-powered crop disease detection system.
I want to build an AI-powered exam proctoring system.

Prompt Evaluation Workflow

The system prompt is isolated in prompts/system_prompt.md. You can freely evaluate and refine the instructions in this file without modifying any application logic. Ensure that changes in the prompt still instruct the model to produce the 7 required reasoning stages matching the Pydantic models. This project is verified against prompt_checker.md for explicit reasoning instructions, internal self-checks, and fallback plans.

Example Logs

Welcome to the AI Project Planner Agent
Enter your project idea: I want to build an AI-powered exam proctoring system.

Connecting to MCP Server...
Reasoning about your project...

Agent decided to research: research_topic(...)

==================================================

Stage – Problem Understanding

Goal
Understand the user's project

Analysis
AI-powered exam monitoring.

Reasoning Summary
Computer vision + authentication.

Reasoning Type
analytical

Self Check
Did I assume the user has access to a webcam? Yes, need to clarify.

Status
Completed

==================================================

... (Repeat for all 7 stages) ...

==================================================

MISSION COMPLETE

Available Tools

1 tool
research_topicA
Search the web for information regarding a specific topic.
Use this tool when you need up-to-date information, technical details, or domain knowledge that you don't possess.

Args:
    query: The search query.
    
Returns:
    A string containing the search results summarized.

Inputs: query (required). Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.8/5.0

    Scored across 1 tool

    Disambiguation5/5

    Only one tool exists, so there is no possibility of confusion between tools. The tool's purpose is clearly defined and distinct.

    Naming Consistency5/5

    The single tool name follows a clear verb_noun pattern and is unambiguous. There are no conflicting conventions.

    Tool Count2/5

    With only one tool, the server is far too thin for the stated purpose of an AI project planner, which would require multiple tools for planning and management.

    Completeness1/5

    The server only supports research, leaving out all planning, task management, and workflow tools that would be essential for an AI project planner agent.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers