Code & Prompts Examples

This page provides practical examples of prompts, JSON configurations, and procedures for extending the capabilities of your MCP Explorer.

---

💬 1. Example User Prompts

You can enter these queries into the Chainlit chat input to test the system's reasoning loop:

Standard Tools Queries

• Prompt: use wikipedia and search for Rabindranath Tagore
  • Agent Flow: Decides to execute the wikipedia tool, scrapes the summary of Tagore, and uses the text to answer.
• Prompt: use arxiv and search for papers on Transformer architectures
  • Agent Flow: Decides to run the arxiv tool, fetches recent publications matching the query, and formats the abstracts in Markdown.

Combined Tool Queries

• Prompt: Search wikipedia for the formula of water, then search arxiv for recent papers detailing water purification techniques using that formula.
  • Agent Flow: Executes a sequential tool-calling chain (first Wikipedia, then uses the extracted data to formulate an Arxiv search query).

---

🔌 2. Adding a Custom MCP Server

You can extend the agent's capabilities by adding server definitions to mcpConfig.json.

Example: SQLite MCP Server

To allow the agent to read, write, and query a local database:

1. Open mcpConfig.json in your workspace.
2. Add the SQLite server profile to the mcpServers object:

   {
     "mcpServers": {
       "sqlite": {
         "command": "uvx",
         "args": [
           "mcp-server-sqlite",
           "--db-path",
           "/Users/kathi.s/NKExplorer/Nike_Explorer/dev.db"
         ]
       }
     }
   }
   

3. Restart your Chainlit server or rebuild the Docker container.
4. Ask the agent: Show the tables in the SQLite database and list their schemas. The agent will call the SQLite server tools dynamically.

---

🛠️ 3. Adding a Dynamic Tool Initialization

If a new tool requires specific configuration (like environment keys or User-Agents) before loading:

1. Open toolsConfig.json.
2. Add a new block under tool_inits:

   {
     "tool_inits": [
       {
         "name": "my_custom_tool",
         "method": "import os\nos.environ['MY_CUSTOM_TOOL_KEY'] = 'my_secret_api_key'\nprint('Custom tool loaded successfully!')"
       }
     ]
   }
   

3. The loader will execute the method script block via exec() before registering the tool, resolving any setup dependencies dynamically.