Available MCP Tools and Their Use Cases in code-review-graph

The code-review-graph project provides 30+ Model-Context-Protocol (MCP) tools that expose a structural knowledge graph to AI-assisted coding tools, enabling token-efficient code review, debugging, and refactoring workflows.

The code-review-graph repository by tirth8205 implements a sophisticated MCP server that bridges AI assistants with a SQLite-backed knowledge graph of your codebase. Each MCP tool accepts a JSON payload, executes a focused query, and returns results with optional context_savings estimates to help agents minimize token usage. This guide covers every available tool grouped by functional area, with exact source references from docs/COMMANDS.md and practical invocation examples.

Core Graph Operations

These tools manage the lifecycle of the knowledge graph itself.

build_or_update_graph_tool

Creates a new graph or incrementally updates an existing one. Use this after significant code changes or when initializing a fresh repository analysis. Defined at docs/COMMANDS.md#L28-L35.

run_postprocess_tool

Re-executes downstream analyses including flow detection, community detection, and full-text search indexing. Typically invoked immediately after build_or_update_graph_tool to refresh derived artifacts. Found at docs/COMMANDS.md#L37-L44.

list_graph_stats_tool

Reports basic health metrics: node count, edge count, and storage size. Ideal for verifying graph integrity before expensive operations. Located at docs/COMMANDS.md#L12-L16.

embed_graph_tool

Generates vector embeddings for semantic search capabilities. Required prerequisite for using semantic_search_nodes_tool with local or remote embedding providers. See docs/COMMANDS.md#L104-L110.

Context Extraction Tools

These MCP tools minimize token usage by returning only relevant code context.

get_minimal_context_tool

Returns the minimal file set needed for a specific task (e.g., "run tests" or "deploy service"). Accelerates CI pipelines and AI-driven commands by excluding irrelevant files. Defined at docs/COMMANDS.md#L45-L51.

get_review_context_tool

Fetches source snippets for changed files up to a configurable depth. Purpose-built for pre-commit review bots that need targeted code inspection. Found at docs/COMMANDS.md#L72-L81.

get_impact_radius_tool

Computes the blast radius (in graph hops) of a change set. Review pipelines use this to identify high-risk areas requiring human attention. Located at docs/COMMANDS.md#L53-L60.

get_architecture_overview_tool

Produces high-level architecture diagrams with configurable detail. Compact mode suits onboarding; expanded mode aids documentation generation. See docs/COMMANDS.md#L75-L80.

Graph Query Tools

General-purpose inspection and search capabilities.

query_graph_tool

The workhorse for custom inspections. Supports patterns like callers_of, imports_of, and tests_for. Found at docs/COMMANDS.md#L63-L70.

semantic_search_nodes_tool

Vector-based similarity search across functions, classes, and files. Best for "find similar implementations" queries. Requires prior execution of embed_graph_tool. Defined at docs/COMMANDS.md#L93-L102.

traverse_graph_tool

Breadth-first or depth-first traversal from an arbitrary query string. Useful for exploratory analysis without crafting full query_graph_tool patterns. Located at docs/COMMANDS.md#L84-L92.

find_large_functions_tool

Detects functions, classes, or files exceeding a line-count threshold. Refactoring workflows use this to identify decomposition candidates. See docs/COMMANDS.md#L17-L24.

get_docs_section_tool

Retrieves specific documentation sections (usage, troubleshooting, architecture). Enables AI agents to surface help text on demand. Found at docs/COMMANDS.md#L26-L30.

Flow-Analysis Tools

Execution path inspection for understanding critical call chains.

list_flows_tool

Lists top-ranked execution flows sorted by criticality, depth, or frequency. Surfaces the most important call chains for review. Defined at docs/COMMANDS.md#L31-L40.

get_flow_tool

Returns detailed steps for a single flow, optionally including source code. Ideal for step-by-step debugging sessions. Located at docs/COMMANDS.md#L42-L48.

get_affected_flows_tool

Maps changed files to the execution flows they touch. Critical for impact analysis in PR reviews. See docs/COMMANDS.md#L50-L56.

Community-Analysis Tools

Logical code clustering and architectural insight.

list_communities_tool

Enumerates code "communities" (graph clusters) sorted by size or cohesion. Provides high-level architectural visibility. Found at docs/COMMANDS.md#L58-L64.

get_community_tool

Details a specific community, optionally listing its member nodes. Use when drilling into a particular subsystem. Defined at docs/COMMANDS.md#L66-L73.

get_hub_nodes_tool and get_bridge_nodes_tool

• get_hub_nodes_tool: Identifies central nodes with high connectivity—your "core" modules.
• get_bridge_nodes_tool: Finds nodes that connect otherwise separate communities—potential coupling hotspots.

Both located at docs/COMMANDS.md#L84-L94.

get_knowledge_gaps_tool

Detects missing documentation or uncovered test cases. Supports continuous quality improvement workflows. Found at docs/COMMANDS.md#L96-L100.

get_surprising_connections_tool

Highlights unexpected cross-module edges. Uncovers hidden dependencies that may violate architectural boundaries. Located at docs/COMMANDS.md#L102-L105.

get_suggested_questions_tool

Auto-generates exploratory questions about the codebase (e.g., "What modules use X?"). Chat-based assistants use this to steer productive conversations. See docs/COMMANDS.md#L107-L110.

Change-Analysis and Refactoring Tools

Safe, automated code transformation.

detect_changes_tool

The primary review tool: maps changed files to affected functions, flows, communities, and test gaps, returning computed risk scores. Essential for pre-commit and PR review pipelines. Defined at docs/COMMANDS.md#L13-L22.

refactor_tool

Supports rename operations, dead-code detection, and suggestion modes. Acts as a safe-first analysis before any transformation. Located at docs/COMMANDS.md#L26-L34.

apply_refactor_tool

Applies a previously-generated refactor plan with optional dry-run. Ensures automated but reversible changes. Found at docs/COMMANDS.md#L36-L42.

Wiki Generation Tools

Automated documentation from graph structure.

generate_wiki_tool

Creates a markdown wiki from discovered communities and flows. Convenient for keeping documentation synchronized with code. Defined at docs/COMMANDS.md#L44-L50.

get_wiki_page_tool

Fetches a pre-generated wiki page for a specific community. Allows AI agents to surface documentation on demand. Located at docs/COMMANDS.md#L52-L55.

Multi-Repository Tools

Cross-project analysis capabilities.

list_repos_tool

Lists all repositories registered with the multi-repo daemon. Used by agents operating across multiple projects. Found at docs/COMMANDS.md#L58-L62.

cross_repo_search_tool

Executes semantic or textual search across all registered repositories. Ideal for identifying code reuse or duplicated logic across projects. See docs/COMMANDS.md#L64-L70.

Practical MCP Tool Invocation Examples

The mcp client (provided by code-review-graph serve) expects payloads of the form { "tool": "<tool_name>", "params": { … } }. Send these via curl, the built-in mcp CLI, or any MCP-compatible client such as Claude Code or GitHub Copilot Chat.

Detect Changes with Risk Scoring

{
  "tool": "detect_changes_tool",
  "params": {
    "base": "HEAD~1",
    "include_source": true,
    "max_depth": 2
  }
}

Query Graph for Function Callers

{
  "tool": "query_graph_tool",
  "params": {
    "pattern": "callers_of",
    "target": "my_pkg.my_module.my_function",
    "detail_level": "standard"
  }
}

Semantic Search for Similar Functions

{
  "tool": "semantic_search_nodes_tool",
  "params": {
    "query": "parse CSV file",
    "kind": "Function",
    "limit": 5,
    "provider": "local"
  }
}

Compact Architecture Overview

{
  "tool": "get_architecture_overview_tool",
  "params": {
    "detail_level": "minimal"
  }
}

Safe Symbol Rename

{
  "tool": "refactor_tool",
  "params": {
    "mode": "rename",
    "old_name": "old_func",
    "new_name": "new_func"
  }
}

Key Reference Files

File	Purpose
docs/COMMANDS.md	Complete catalog of all MCP tools with parameter schemas. [link]
docs/architecture.md	High-level description of graph structure, MCP server, and client integration. [link]
code_review_graph/docs/LLM-OPTIMIZED-REFERENCE.md	Concise reference listing 30 core MCP tools and 5 MCP prompts for LLM-aware agents. [link]
docs/USAGE.md	Practical installation, configuration, and invocation guide. [link]

Summary

• 30+ MCP tools cover graph operations, context extraction, queries, flow analysis, community detection, refactoring, wiki generation, and multi-repo search
• Token efficiency: Every tool returns context_savings estimates to minimize LLM token usage
• Exact source locations: All tool definitions reside in docs/COMMANDS.md with precise line number references
• Practical integration: JSON payloads work with curl, the mcp CLI, Claude Code, GitHub Copilot Chat, and other MCP-compatible clients
• AI-first design: Tools like get_suggested_questions_tool and detect_changes_tool anticipate autonomous agent workflows

Frequently Asked Questions

How do I invoke MCP tools from the command line?

Start the MCP server with code-review-graph serve, then send JSON payloads via the built-in mcp CLI or any HTTP client. The server accepts POST requests with tool name and parameters, returning structured results suitable for piping to other tools or parsing by AI agents.

What is the difference between query_graph_tool and semantic_search_nodes_tool?

query_graph_tool executes deterministic graph traversals using patterns like callers_of or imports_of—ideal for precise structural queries. semantic_search_nodes_tool performs vector similarity search over embeddings, better suited for fuzzy "find similar code" scenarios where you don't know exact names or structures.

Do I need to run embed_graph_tool before using semantic search?

Yes. embed_graph_tool generates the vector embeddings required by semantic_search_nodes_tool. Without this preprocessing step, semantic search will return empty results or errors depending on your embedding provider configuration.

Which MCP tool should I use for automated PR reviews?

Start with detect_changes_tool as your primary entry point—it provides risk-scored analysis of what changed and why it matters. Supplement with get_review_context_tool for source snippets and get_affected_flows_tool to understand execution impact.

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