The Nuanced MCP Server can be easily integrated with to provide call graph analysis capabilities directly in your conversations.

​
Installation

​
Prerequisites

Before configuring , ensure you’ve completed the installation steps
  1. Configure npm authentication with your license key
  2. Install the MCP server globally

​
Configuration

To add the Nuanced MCP server to Amp, run this command in your terminal before starting Amp: 
amp mcp add nuanced -- npx -y @nuanced-dev/nuanced-mcp-ts@latest
This will configure Amp to use the latest nuanced-mcp-ts package as an MCP server named “nuanced”.

​
Best Practices

The Nuanced MCP Server uses static analysis to provide a rich map of your project. This reduces hallucinations and waste while improving output. Use these guidelines to get the most out of Nuanced:

​
Example Use Cases

  1. Understanding Code — Use the structured map to see how functions relate.
    “Use Nuanced on process_order to explain how this function connects to other parts of the system.”
  2. Refactoring Code — Identify dependencies before making changes.
    “Use Nuanced on calculate_total. Suggest helpers to extract based on its call graph.”
  3. Adding Tests — Surface edge cases and dependencies for better coverage.
    “Use Nuanced on authenticate_user to generate unit tests along call paths.”
  4. Adding New Features — Plan where to extend existing code.
    “Starting with send_invoice, use Nuanced to plan a new payment method.”
The Nuanced MCP Server works best when you specify “jump-in points” by providing function names.

​
Steering

By default, may not use the Nuanced MCP Server for code context unless you explicitly prompt it to “use Nuanced”. To ensure consistently leverages the Nuanced MCP Server, we recommend adding or modifying your project’s file with the following: 
## Code understanding

Whenever you need to retrieve or understand code context, start with the Nuanced MCP Server. Use the returned metadata to map the call graph before diving into the source.

- Treat each node as a function. `callers` are upstream locations (who invokes it) and `callees` are downstream dependencies (who it invokes). Use this to sketch the subgraph around your target.
- Follow call chains while they add insight or the next hop no longer informs your task. Skip <native> callees; they’re leaf nodes.
- After reviewing the metadata, read the function source using the provided start/end ranges to confirm behavior, understand data flow, and capture nuances the graph can’t show.
- Combine structural findings from the graph with behavioral details from the source in your response so the user gets both relationship context and concrete implementation notes.

​
Next Steps

  • Explore the full feature set in the Overview
  • Try the MCP server with Claude