Skip to Content
DocumentationTools ReferenceOverview

MCP Tools Overview

dev-agent provides five tools through the Model Context Protocol (MCP). These tools give AI assistants deep understanding of your codebase.

Available Tools

ToolPurpose
dev_searchSemantic code search with snippets
dev_refsQuery code relationships (callers/callees)
dev_mapCodebase structure overview with change frequency
dev_patternsFile pattern analysis (finds similar code, compares 5 pattern categories)
dev_statusRepository indexing status and health checks

New in v0.8.5

  • dev_patterns β€” Refactored for comprehensive pattern analysis (5 categories)
  • Performance β€” 5-10x faster pattern analysis via batch scanning (500-1000ms)
  • Accuracy β€” Semantic similarity using document embeddings, extension filtering
  • Simplified API β€” Streamlined interface (no action parameter needed)

v0.5.0

  • Enhanced indexing β€” Arrow functions, React hooks, and exported constants now extracted
  • dev_search β€” Better coverage of modern JavaScript patterns (hooks, utilities, configs)

v0.4.0

  • dev_map β€” Change frequency indicators (πŸ”₯ hot, ✏️ active, πŸ“ recent)

v0.3.0

  • dev_refs β€” Query what calls what and what is called by what
  • dev_map β€” Get codebase overview with hot paths and smart depth

How Tools Are Exposed

Tools are exposed via the Model Context Protocol (MCP), a standard for AI tool integration:

AI Assistant (Cursor/Claude) β”‚ β”‚ MCP Protocol (stdio) β–Ό dev-agent MCP Server β”‚ β”œβ”€β”€ SearchAdapter β†’ dev_search β”œβ”€β”€ RefsAdapter β†’ dev_refs β”œβ”€β”€ MapAdapter β†’ dev_map β”œβ”€β”€ InspectAdapter β†’ dev_patterns └── StatusAdapter β†’ dev_status (includes health checks)

Tool Response Format

All tools return structured responses with metadata:

{ "success": true, "data": { "query": "authentication", "content": "1. [89%] class: AuthMiddleware..." }, "metadata": { "tokens": 245, "duration_ms": 123, "results_total": 15, "results_returned": 10 } }

The metadata field helps AI assistants understand:

  • Token cost of the response
  • Performance metrics
  • Result completeness (truncated or full)

Output Formats

Most tools support two output formats:

  • compact (default) β€” Token-efficient, markdown output
  • verbose β€” Full JSON with all details

Next Steps

Last updated on