MCP Glootie - Agentic Coding Optimizer

An MCP tool that improves the output and wall clock problem solving performance of your programming agent.

Glootie extrapolates on a senior developer's preferred workflows and diagnostic processes, making that functionality available to agents. The thinking is if 50% of a senior developer's coding effort is spent on this tool, then many best practices get captured in a way that other developers can pick up on and hopefully improve.

What Glootie DOES Do

Execute code first before editing - You won't believe the advantage you get from just this. It encourages agents to hypothesize and test code before editing files, grounding edits in truth. Execute in the repo with available libraries for Node, Deno, and Bash.

Semantic code searching - Fast, compatible semantic search embedded in Glootie. No need for third-party code searches.

Surgical code updates with AST-grep - AST functionality is native to AI these days. Huge performance boost from access to it.

Internal batch capabilities - Trade multi-turn latency for bundling, drastically reducing turnaround speed where providers impose delays.

Built-in step by step reasoning - Optimized for daily use and token reduction. Agents can assign reasoning processes to project folders, enabling intelligent cross-repo work.

What Glootie DOESN'T Do

Glootie is not a product - it's an in-house programming tool for an independent developer. You can receive the same benefits without making the same tools, but there's no company and there's no service.

Tools (15 Total)

Execution: executenodejs, executedeno, executebash

Search & Analysis: searchcode, astgrep_search, astgrep_replace, astgrep_lint

Project Intelligence: project_analyze, file_navigator, dependency_analyzer, performance_profiler, quality_analyzer

Utilities: retrieve_overflow, batch_execute, sequentialthinking

Example Agent Prompts

These prompts demonstrate how agents can leverage MCP REPL's full capabilities:

Code Analysis & Refactoring

Use MCP REPL to:
1. Search for all functions that handle user authentication in this codebase
2. Analyze the current implementation patterns using semantic search
3. Execute test cases to understand current behavior
4. Propose and test refactored implementations
5. Use batch execution to validate all changes work together

Feature Development Workflow

Develop a new caching system using MCP REPL:
1. Use searchcode to find existing caching patterns in the codebase
2. Execute prototype implementations with executenodejs to test concepts
3. Use astgrep_search to find integration points
4. Implement and test the feature using batch execution
5. Use sequentialthinking to track design decisions and trade-offs

Performance Optimization

Optimize database queries using MCP REPL:
1. Search for all database query functions with semantic search
2. Execute performance benchmarks using executenodejs
3. Use astgrep_analyze to understand query patterns
4. Test optimized implementations
5. Use batch_execute to run comparative performance tests
6. Document improvements using sequentialthinking

Performance Analysis Results

Based on comprehensive testing comparing Claude Code with and without MCP REPL tools:

Key Metrics

Context Optimization: 89% reduction in redundant text (3,240 → 360 characters)
Token Efficiency: 810 tokens saved per context window
Tool Performance: <1ms load time (exceeds 50ms industry target)
Memory Usage: 4.46MB (exceeds 100MB target)

Analysis Style Comparison

With MCP REPL: Concise, focused analysis with exact file paths and line numbers
Standard Claude: Comprehensive coverage with structured documentation and roadmaps
Both Approaches: Complementary - serve different analysis needs

Real-World Test Results

Security audit task on complex Node.js application:

Standard Claude: 246 lines of comprehensive analysis with implementation roadmap
MCP REPL: 58 lines of focused technical analysis with precise code locations
Value: Each approach serves different use cases effectively

For detailed performance analysis, see claude-test-dir/performance-analysis.md

Code Quality & Security

Conduct a security audit using MCP REPL:
1. Search for potential security vulnerabilities with semantic search
2. Use astgrep_lint to validate against security rules
3. Execute penetration test scripts
4. Use batch_execute to run comprehensive security scans
5. Generate remediation reports with structured findings

Multi-Repository Analysis

Analyze architecture across multiple repositories:
1. Use searchcode to understand component relationships
2. Execute cross-repo dependency analysis
3. Use sequentialthinking to track architectural decisions
4. Generate integration documentation
5. Test deployment scenarios across repos

Learning & Documentation

Generate comprehensive code documentation:
1. Use semantic search to find all public APIs
2. Execute code to understand behavior and edge cases
3. Use astgrep_analyze to extract function signatures and types
4. Batch execute usage examples
5. Structure findings into documentation using sequentialthinking

Migration & Modernization

Modernize legacy code using MCP REPL:
1. Search for legacy patterns and dependencies
2. Execute compatibility tests
3. Use astgrep_replace for safe transformations
4. Batch test modernized implementations
5. Track migration progress with sequentialthinking

Installation

Claude Code

claude mcp add -s user repl "npx" "-y" "mcp-repl"

Cursor

Add to your Cursor mcpServers.json configuration:

{
  "mcpServers": {
    "mcp-repl": {
      "command": "npx",
      "args": [
        "-y", "mcp-repl"
      ],
      "env": {},
      "disabled": false,
      "autoApprove": [
        "executenodejs",
        "executedeno",
        "executebash",
        "retrieve_overflow",
        "searchcode",
        "astgrep_search",
        "astgrep_replace",
        "astgrep_lint",
        "astgrep_analyze",
        "astgrep_enhanced_search",
        "astgrep_multi_pattern",
        "astgrep_constraint_search",
        "astgrep_project_init",
        "astgrep_project_scan",
        "astgrep_test",
        "astgrep_validate_rules",
        "astgrep_debug_rule",
        "batch_execute",
        "sequentialthinking"
      ]
    }
  }
}

GitHub Copilot

Add to your GitHub Copilot mcpServers.json configuration:

{
  "mcpServers": {
    "repl": {
      "command": "npx",
      "args": ["-y", "@anentrypoint/mcp-repl"],
      "env": {},
      "type": "local",
      "tools": [
        "executenodejs",
        "executedeno",
        "executebash",
        "retrieve_overflow",
        "searchcode",
        "astgrep_search",
        "astgrep_replace",
        "astgrep_lint",
        "astgrep_analyze",
        "astgrep_enhanced_search",
        "astgrep_multi_pattern",
        "astgrep_constraint_search",
        "astgrep_project_init",
        "astgrep_project_scan",
        "astgrep_test",
        "astgrep_validate_rules",
        "astgrep_debug_rule",
        "batch_execute",
        "sequentialthinking"
      ]
    }
  }
}

VSCode

Add to your VSCode MCP configuration:

{
    "servers": {
        "repl": {
            "command": "node",
            "args": [
                "c:/dev/mcp-repl/src/direct-executor-server.js"
            ],
            "env": {},
            "type": "stdio"
        }
    },
    "inputs": []
}

Dependencies

@modelcontextprotocol/sdk (^1.11.0) - MCP SDK
@xenova/transformers (^2.17.2) - Semantic search
@ast-grep/napi (^0.28.0) - AST analysis
ignore (^7.0.5) - .gitignore handling

Requirements: Node.js ≥16.0.0, 50MB+ memory, 100MB+ disk

Contributing

Performance first: Maintain efficiency standards
Token efficiency: Tool descriptions <500 chars
MCP compliance: Follow protocol best practices
Testing: Full coverage with benchmarks

License

MIT License

MCP Glootie - The most undervalued MCP tool. Let me know what you think of Glootie's performance.

MCP Glootie - Agentic Coding Optimizer