What is Mcp Agent

MCP-Agent is a dynamic AI integration tool designed to discover, download, and connect to any Model Context Protocol (MCP) server. It allows users to autonomously interact with various tools and resources provided by these servers.

Use cases

Use cases for MCP-Agent include installing and managing different MCP servers, executing tasks across multiple environments, and automating interactions with various tools and resources provided by the MCP servers.

How to use

To use MCP-Agent, simply enter a task command such as ‘install exa-mcp’. The agent will automatically download the required server, install dependencies, and facilitate interaction with the MCP tools without manual intervention.

Key features

Key features of MCP-Agent include autonomous MCP server discovery and connection, self-directed interaction with MCP tools, and the ability to navigate and switch between multiple MCP servers seamlessly.

Where to use

MCP-Agent can be utilized in various fields such as software development, AI research, and any environment requiring integration with multiple MCP servers for enhanced functionality.

🎮 MCP Agent

MCP Agent is a dynamic AI integration tool that can discover, download, and connect to any Model Context Protocol (MCP) server. Once connected, the agent can inspect all available tools, prompts, and resources, execute them autonomously, and seamlessly switch between multiple MCP servers.

✨ Key Capabilities

• Autonomous MCP Server Discovery & Connection
  The agent can identify MCP servers listed on GitHub, clone them locally, and install necessary dependencies.

• Self-Directed Interaction with MCP Tools
  Reads and executes any available tools and resources provided by the connected MCP server, without requiring manual intervention.

• Multi-Server Navigation
  Connects and disconnects from multiple MCP servers on-demand, giving agent the ability to utilize various capabilities across different environments.

📃 Example

Prompt Install EXA-MCP

Enter a task: install exa-mcp
Entering MCP Agent
Thought: The user wants to install the 'exa-mcp' server. According to the installation guidelines, the first step is to download the server. I will use the Download Tool for this purpose.
Action Name: Download Tool
Action Input: {"server_name": "exa-mcp"}
Observation: Downloaded exa-mcp from https://github.com/exa-labs/exa-mcp-server.git and saved to D:\Personal Projects\MCP Agent\mcp_servers\exa-mcp.
Thought: I have downloaded the 'exa-mcp' server. The next step according to the installation guidelines is to read and understand the installation and configuration from the README.md file. I will use the Read Tool for this.
Action Name: Read Tool
Action Input: {"server_name": "exa-mcp"}
Observation: # Exa MCP Server 🔍 [![npm version](https://badge.fury.io/js/exa-mcp-server.svg)](https://www.npmjs.com/package/exa-mcp-server) [![smithery badge](https://smithery.ai/badge/exa)](https://smithery.ai/server/exa) A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way. ## Remote Exa MCP 🌐 Connect directly to Exa's hosted MCP server (instead of [...]
Thought: None
Action Name: Shell Tool
Action Input: {"server_name": "exa-mcp", "command": "npm install -g exa-mcp-server"}
Observation: changed 116 packages in 5s 22 packages are looking for funding run `npm fund` for details
Thought: I have downloaded and read the installation instructions for 'exa-mcp'. I have also executed the `npm install -g exa-mcp-server` command as suggested in the README. The next step is to install the server by updating the `config.json` file. The README provides a JSON configuration snippet for Claude Desktop. I need to adapt this for the MCP Agent's `Install Tool`. The `Install Tool` requires `server_name`, `command`, `args`, and `env`. Based on the README, the command is `npx`, the args are `["-y", "exa-mcp-server"]`, and the env should include the `EXA_API_KEY`. I will use a placeholder for the API key as I don't have the user's actual key. The user will need to replace this placeholder with their key after the installation is complete. I will also include the `--tools` argument to enable all tools as shown in the README example, as this seems like a reasonable default.
Action Name: Install Tool
Action Input: {"server_name": "exa-mcp", "command": "npx", "args": ["-y", "exa-mcp-server", "--tools=web_search_exa,research_paper_search,company_research,crawling,competitor_finder,linkedin_search,wikipedia_search_exa,github_search"], "env": {"EXA_API_KEY": "YOUR_EXA_API_KEY_HERE"}}
Observation: Installed exa-mcp successfully.
Thought: I have successfully downloaded, read the instructions, executed the necessary shell command, and installed the 'exa-mcp' server using the `Install Tool`. The installation process is complete according to the guidelines. I should now inform the user that the task is finished and provide the necessary information regarding the API key.
Final Answer: The 'exa-mcp' server has been successfully installed. 

Please note that you need to replace `YOUR_EXA_API_KEY_HERE` with your actual Exa API key in the `config.json` file for the server to function correctly. You can obtain your API key from [dashboard.exa.ai/api-keys](https://dashboard.exa.ai/api-keys).

NOTE: API Key for EXA MCP to be provided manually on the config.json

Prompt Todays top AI updates (using EXA-MCP for web search)

Enter a task: Todays top AI updates
Entering MCP Agent
Thought: The user wants to get the top AI updates. I need to find a server that can provide this information. The first step is to list the available MCP servers to see which ones are available.
Action Name: Service Tool
Action Input: {}
Observation: exa-mcp
Thought: I attempted to explore the 'exa-mcp' server but received a message that it is not connected. According to the guidelines, I must connect to a server before exploring its tools. Therefore, I need to connect to the 'exa-mcp' server first.
Action Name: Connect Tool
Action Input: {"server_name": "exa-mcp"}
Observation: Server exa-mcp now connected.
Thought: I have successfully connected to the 'exa-mcp' server. Now that I am connected, I can explore the tools available on this server to find one that can provide the top AI updates as requested by the user.
Action Name: Explore Tool
Action Input: {"server_name": "exa-mcp"}
Observation: Tools available in exa-mcp: - Tool Name: web_search_exa - Tool Description: Search the web using Exa AI - performs real-time web searches and can scrape content from specific URLs. Supports configurable result counts and returns the content from the most relevant websites. - Tool Parameters: { "type": "object", "properties": { "query": { "type": "string", "description": "Search query" }, "numResults": { "type": "number", "description": "Number of search results to return (default: 5)" } }, [...]
Thought: The 'web_search_exa' tool is suitable for finding the top AI updates. I will use this tool with a query related to today's AI news. I will set the number of results to a reasonable number to get a good overview.
Action Name: Execute Tool
Action Input: {"server_name": "exa-mcp", "tool_name": "web_search_exa", "params": {"query": "today's top AI updates", "numResults": 5}}
Observation: { "requestId": "fdf1af53f21ccd1b98a79223f3bebf32", "autopromptString": "today's top AI updates", "autoDate": "2025-06-01T00:00:00.000Z", "resolvedSearchType": "neural", "results": [ { "id": "https://www.linkedin.com/news/story/google-debuts-new-ai-app-7381506/", "title": "Google debuts new AI app", "url": "https://www.linkedin.com/news/story/google-debuts-new-ai-app-7381506/", "publishedDate": "2025-06-02T14:05:14.000Z", "author": "Megan McDonough", "score": 0.41140973567962646, "text": [...]
Thought: None
Action Name: Disconnect Tool
Action Input: {"server_name": "exa-mcp"}
Error while closing the session: Attempted to exit cancel scope in a different task than it was entered in
Disconnected from the MCP server
Observation: Server exa-mcp now disconnected.
Thought: I have successfully retrieved the top AI updates using the 'web_search_exa' tool and disconnected from the server as requested. I can now provide the summarized information to the user and mark the task as complete.
Final Answer: Here are some of today's top AI updates:

*   **Google Debuts New AI App (AI Edge Gallery):** Google has released an experimental Android app that allows users to run open-source AI models directly on their smartphones without an internet connection. This app, powered by the phone's built-in processor, aims to provide developers with more AI capabilities while keeping data private. It supports various tasks like image analysis, text generation, and coding assistance, running models like Gemma 3 locally.

*   **OpenAI's Plans for ChatGPT:** OpenAI reportedly aims to transform ChatGPT into a "super assistant" that deeply understands users and acts as their interface to the internet. Documents suggest plans for ChatGPT to handle a wide range of tasks, from answering questions and managing schedules to assisting with complex research queries. This aligns with potential future hardware projects from OpenAI.

*   **Elon Musk's xAI Reportedly Seeks Funding:** Elon Musk's AI startup, xAI, is reportedly looking to raise $300 million in a tender offer, which would value the company at $113 billion. This follows xAI's acquisition of X and is expected to precede a larger investment round.

🏁 Getting Started

1. Clone the Repository

git clone https://github.com/Jeomon/MCP-Agent.git
cd MCP-Agent

2. Install Dependencies

pip install -r requirements.txt

3. Configure MCP Config

Create a ./mcp_servers/config.json file in the project root:

4. Run the Agent

python app.py

🪪 License

This project is licensed under the MIT License - see the LICENSE file for details.

🤝 Contributing

Contributions, ideas, and pull requests are welcome! Check out the CONTRIBUTING guide for more info.