Skip to content

NEXT_AGENT_PROMPT.md

Latest commit

 

History

History
122 lines (97 loc) · 4 KB

NEXT_AGENT_PROMPT.md

File metadata and controls

122 lines (97 loc) · 4 KB

Next Agent: Research & Document CodeMode Ecosystem

Context

You're working on CodeMode Unified - a secure code execution platform with MCP (Model Context Protocol) integration. We just successfully implemented:

  1. Health check security (masked env vars)
  2. Fixed MCP execute endpoint to call tools directly
  3. Added verbose tool discovery logging
  4. Verified 6 MCP servers connected (37 total tools)

Your Mission

Phase 1: Research CodeMode Ecosystem

Research and analyze what other packages in the ecosystem are doing:

Primary Research Targets:

  1. CF's CodeMode (Cloudflare Workers version)

    • Architecture and approach
    • Security model
    • MCP integration patterns

  2. Open Source Alternatives

    • E2B (code execution platform)
    • Replit's Pyodide integration
    • WebContainer (StackBlitz)
    • Other sandboxed execution platforms

  3. Community Sentiment

    • Reddit discussions about code execution platforms
    • Hacker News threads on MCP and code execution
    • YCombinator discussions
    • Developer feedback and pain points

Research Questions:

  • What are developers saying about CodeMode alternatives?
  • What features are most valued?
  • What are the main complaints/limitations?
  • What security approaches are working well?
  • How are others handling MCP tool integration?
  • What's the state of TypeScript generation from MCP schemas?

Phase 2: Document Key Findings

Based on research, create comprehensive documentation covering:

  1. COMPETITIVE_ANALYSIS.md

    • Feature comparison matrix
    • Security model comparison
    • Performance benchmarks (if available)
    • Developer experience comparison

  2. MCP_AUTO_GENERATION_SPEC.md

    • Research: How others are generating types from MCP tools
    • Design: Our approach to auto-generating TypeScript from tool schemas
    • Implementation plan for:
      • Tool discovery → TypeScript interface generation
      • Parameter validation types
      • Return type inference
      • Auto-generated documentation
      • IDE autocomplete support

  3. ROADMAP.md

    • Feature priorities based on research
    • Gaps we can fill that others haven't
    • Unique value propositions

Key Files to Review

Current Implementation:

  • src/mcp/aggregator.ts - Tool discovery with verbose logging (lines 213-244)
  • src/server.ts - MCP execute endpoint (lines 196-247)
  • .mcp.json - 6 configured servers including claude-code MCP

Tool Discovery Output: Tool discovery now logs:

🔍 [TOOL DISCOVERY] Server: helpscout
📋 [TOOL DISCOVERY] Found 8 tools
  ✨ [TOOL] Original: "searchInboxes" → Namespace: "helpscout.searchInboxes"
     📝 Description: STEP 1: Always use this FIRST...
     📥 Parameters: [query, limit, cursor]

Recent Output: Check /tmp/codemode-verbose.log for complete tool discovery with:

  • automem (7 tools)
  • sequential-thinking (1 tool)
  • context7 (2 tools)
  • WordPressAPI (12 tools)
  • helpscout (8 tools)
  • claude-code (15 tools)

Success Criteria

  1. ✅ Comprehensive competitive analysis
  2. ✅ Clear understanding of community needs/pain points
  3. ✅ Detailed spec for TypeScript auto-generation from MCP schemas
  4. ✅ Prioritized roadmap based on research findings
  5. ✅ Identification of unique positioning opportunities

Resources

Research Starting Points:

  • Search: "CodeMode Cloudflare" reddit
  • Search: "MCP Model Context Protocol" hacker news
  • Search: "code execution sandbox" ycombinator
  • Search: "E2B code interpreter" comparison
  • GitHub: Search for MCP TypeScript generators
  • NPM: Search for MCP-related packages

Technical Context:

  • MCP SDK: @modelcontextprotocol/sdk
  • Our servers use stdio transport
  • Tool schemas include inputSchema (JSON Schema format)
  • We need: TypeScript interfaces + Zod schemas + docs

Output Format

Create markdown documentation files in the project root following the established pattern (see HEALTH_CHECK_FIX.md, MCP_EXECUTE_FIX.md for style reference).

Start with: Research phase, then document findings with actionable insights for implementation.