How to Get Code Snippets Using Codebase-Memory-MCP: A Complete Guide

Use the get_code_snippet MCP tool by passing a qualified name (e.g., myproject.module.Class.method) discovered via search_graph to retrieve the original source lines from the indexed repository.

The DeusData/codebase-memory-mcp repository provides a Model Context Protocol (MCP) server that indexes your codebase into a knowledge graph, enabling precise source code retrieval. To get code snippets using Codebase-Memory-MCP, you must understand how to construct qualified names and invoke the dedicated get_code_snippet tool that queries the underlying SQLite graph storage.

Understanding the get_code_snippet Tool

According to the source code in src/mcp/mcp.c (line 4139), the handle_get_code_snippet function implements the core logic for this MCP tool. The tool expects a qualified name—a dot-delimited path that uniquely identifies a symbol in the indexed graph (e.g., myproject.services.order.OrderProcessor.process_order).

Internally, the implementation:

• Queries the SQLite knowledge graph table cbm_node_t to locate the node
• Copies the node to a temporary heap allocation via copy_node
• Resolves the absolute file path and verifies it resides within the indexed repository boundary through resolve_snippet_source
• Reads the requested line range and returns a UTF-8 sanitized JSON payload containing the source text and metadata

The user-facing documentation for this tool is generated in src/cli/cli.c around line 486.

Prerequisites: Finding the Qualified Name

Before retrieving snippets, you must discover the exact qualified name. Use the search_graph or search_code tools to locate symbols.

Example workflow:

1. Search for functions matching a pattern
2. Extract the qualified_name from the results
3. Pass it to get_code_snippet

Three Ways to Retrieve Code Snippets

Method 1: CLI (Command Line Interface)

The CLI provides a straightforward interface for single-shot queries. First, search for the symbol, then retrieve its source.

# Step 1: Find the qualified name

codebase-memory-mcp cli search_graph '{"label":"Function","name_pattern":".*process.*"}'

# Step 2: Retrieve the source snippet

codebase-memory-mcp cli get_code_snippet '{"qualified_name":"myproject.services.order.OrderProcessor.process_order"}'

The CLI translates the JSON-RPC call into a request to the MCP server and returns a structured JSON response:

{
  "status": "ok",
  "qualified_name": "myproject.services.order.OrderProcessor.process_order",
  "file_path": "services/order/order_processor.py",
  "start_line": 42,
  "end_line": 58,
  "source": "def process_order(self, order):\n    # ... implementation ..."

}

Method 2: Direct JSON-RPC Call

For programmatic access, send raw JSON-RPC messages to the MCP server's STDIN/STDOUT pipe (or HTTP bridge if enabled).

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "get_code_snippet",
  "params": {
    "qualified_name": "myproject.services.order.OrderProcessor.process_order"
  }
}

The server invokes handle_get_code_snippet (defined at line 4139 in src/mcp/mcp.c) and returns the snippet metadata and source text.

Method 3: Python Client

The official Python client available on PyPI abstracts the JSON-RPC communication.

from codebase_memory_mcp import MCPClient

client = MCPClient()  # connects to the local MCP server

# Optional: Find the symbol first

results = client.search_graph(label="Function", name_pattern=".*process.*")
qn = results[0]["qualified_name"]

# Retrieve the snippet

snippet = client.get_code_snippet(qualified_name=qn)
print(snippet["source"])

Optional Parameters and Advanced Usage

The get_code_snippet tool supports several optional parameters to refine the output:

• include_neighbors: Set to true to include surrounding function or class definitions in the response
• start_line / end_line: Specify 1-based line numbers to restrict the snippet to a specific range
• raw (CLI flag): Use --raw to output the raw snippet without the JSON wrapper

Example with line range restriction:

codebase-memory-mcp cli get_code_snippet '{"qualified_name":"myproject.utils.helper","start_line":10,"end_line":25}'

Handling Ambiguous Results

When a qualified name matches multiple symbols (e.g., overloaded functions or duplicate names across modules), the tool returns an "ambiguous" status. The response includes a snippet_suggestions array (see implementation at line 4445 in src/mcp/mcp.c) containing possible matches.

Review the suggestions, select the correct qualified_name, and retry the request with the specific identifier.

Summary

• Use get_code_snippet: The primary MCP tool for retrieving source code from the indexed graph
• Require qualified names: Always discover the exact symbol path using search_graph or search_code first
• Three access patterns: CLI for manual queries, JSON-RPC for custom integrations, Python client for application development
• Core implementation: Located in src/mcp/mcp.c (handle_get_code_snippet, resolve_snippet_source)
• Handle ambiguity: Watch for "ambiguous" status and use snippet_suggestions to disambiguate targets

Frequently Asked Questions

What is a qualified name in Codebase-Memory-MCP?

A qualified name is a dot-delimited identifier that uniquely represents a symbol in the codebase knowledge graph (e.g., package.module.Class.method). You must obtain this exact string from search_graph results before calling get_code_snippet, as the tool uses it to query the cbm_node_t table in the underlying SQLite database.

Can I retrieve code snippets from files outside the indexed repository?

No. The resolve_snippet_source function in src/mcp/mcp.c explicitly verifies that the resolved absolute file path resides within the indexed repository boundaries. If the file falls outside the indexed scope, the request will fail as a security measure.

How do I get more context around a specific function?

Set the include_neighbors parameter to true in your request. This returns not only the target symbol but also surrounding function and class definitions, providing broader context without requiring multiple separate queries.

What happens if the qualified name does not exist?

The tool returns an error status indicating the symbol was not found. If the name is ambiguous (matches multiple nodes), you will receive an "ambiguous" status with a snippet_suggestions list containing possible alternatives, allowing you to select the correct qualified name and retry.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}