Practical MCP with FastMCP & LangChain

Step 6 - Integration: Composing the Final Answer

Once the MCP server finishes executing the tool, it sends a JSON-RPC response back to the client. The Client then unwraps this message and hands the raw data to the host application.

The Feedback Loop

The Host does not show this raw JSON to the user immediately. Instead, it performs a context injection step:

1. It appends the tool's result to the LLM's conversation history.
2. It sends a new internal prompt to the LLM, effectively saying:

Here is the data you requested from the 'get_air_quality' tool. 

--
[..RAW JSON DATA..]
--

Now, use this to answer the user's original question.

The LLM then processes this new prompt, extracts the relevant information from the tool's output, and generates a final response for the user. For example:

The air quality in San Francisco is currently **Good**. The AQI is **42**, and PM2.5 levels are around **8 µg/m³**, which is well within the healthy range.

Post-Processing & Application Logic