Analyze Claude Tool Usage

Purpose: Analyze which tools Claude uses most frequently in conversations using the MCP server

---

Using the MCP Tool

The vibe-check MCP server provides a vibe_tools tool that analyzes Claude's tool usage patterns automatically.

Tool: mcp__vibe-check__vibe_tools

Parameters:

• days (optional): Number of days to analyze (default: 30)
• repo (optional): Filter to specific repository
• show_combinations (optional): Include tool combination analysis (default: False)

---

Usage Flow

When user requests tool analysis:

1. Extract filters from user's question:

   • Time range (if specified) → convert to days parameter
   • Repository (if specified) → use repo parameter
   • If they ask about "combinations" or "patterns" → set show_combinations=True

2. Call the MCP tool with appropriate parameters

3. Present results - the tool returns formatted output with:

   • Most used tools with usage counts and percentages
   • Visual bar charts
   • Total tool usage count
   • Tool combinations (if requested)

4. Offer insights based on the data:

   • Which tools dominate
   • What the tool mix suggests about work patterns
   • Recommendations for workflow improvements

---

Example Interactions

Example 1: Overall Tool Usage

User: "what tools do I use most?"

Claude: [Calls mcp__vibe-check__vibe_tools with days=30]

[Tool returns top tools with usage stats and bar charts]

This shows you're doing a lot of file reading and bash commands, suggesting
you're working with existing code and running tests/builds frequently.

Example 2: Recent Tool Usage

User: "what tools have I been using this week?"

Claude: [Calls mcp__vibe-check__vibe_tools with days=7]

[Tool returns last 7 days of tool usage]

Example 3: Repo-Specific Analysis

User: "analyze my tool usage in the vibe-check project"

Claude: [Calls mcp__vibe-check__vibe_tools with repo="vibe-check"]

[Tool returns vibe-check repo tool usage]

Example 4: With Combinations

User: "show me which tools I use together most often"

Claude: [Calls mcp__vibe-check__vibe_tools with days=30, show_combinations=True]

[Tool returns top tools plus common tool combinations like "Grep + Read"]

---

Interpreting Results

Common Tool Patterns

Heavy Read usage: Exploring and understanding code

• Suggests research/learning phase or code review work

Heavy Edit usage: Actively modifying code

• Suggests active development and refactoring

Heavy Bash usage: Running commands, tests, git operations

• Suggests integration work, testing, deployment

Grep + Glob combination: Searching for code patterns

• Suggests investigative work or debugging

Read + Edit combination: Understanding then modifying

• Suggests careful, iterative development

Write usage: Creating new files

• Suggests new feature development or documentation

---

Related Tools

Use these complementary MCP tools:

• mcp__vibe-check__vibe_stats: See overall usage statistics
• mcp__vibe-check__vibe_recent: View recent sessions to see context
• mcp__vibe-check__vibe_search: Search for specific tool usage patterns
• mcp__vibe-check__vibe_session: Deep dive into a specific session's tools

---

Error Handling

The MCP tool handles errors automatically:

• If no tool usage found, it explains that no data is available
• If database not accessible, it provides helpful file path information
• All database locking is handled internally with read-only mode

---

Tips

• Use default 30 days first to get a good overview
• Add show_combinations=True to understand workflow patterns
• Compare tool usage across different repos to see how work differs
• Use tool analysis to identify if you're spending too much time on certain activities
• If you see lots of Bash errors, might indicate test/build issues
• High Grep usage might suggest code organization could be improved

---

Advanced: Custom SQL Queries

For custom tool analysis not covered by vibe_tools, use the mcp__vibe-check__vibe_sql tool:

mcp__vibe-check__vibe_sql(
    query="""
        SELECT json_extract(value, '$.name') as tool, COUNT(*) as uses
        FROM conversation_events, json_each(json_extract(event_data, '$.message.content'))
        WHERE json_extract(value, '$.type') = 'tool_use'
        GROUP BY tool ORDER BY uses DESC
    """,
    limit=100
)

This provides read-only access to the full database with custom SQL. See the vibe-check-sql skill for more examples.

For schema information, see ~/.vibe-check/SCHEMA.md (auto-generated) or use the Read tool.