[Extension]: Add

[capatinore]

Extension ID

contextforge-mcp

Extension Name

ContextForge MCP

Version

0.1.0

Description

Integrates codebase-memory-mcp + headroom into Spec Kit — adds code intelligence and context compression. Reduces token usage 90%+ during implement phase.

Author

capatinore

Repository URL

https://github.com/capatinore/contextforge-mcp

Download URL

https://github.com/capatinore/contextforge-mcp/releases/download/ext-v0.1.0/contextforge-mcp-speckit-extension.zip

License

MIT

Homepage (optional)

No response

Documentation URL (optional)

No response

Changelog URL (optional)

No response

Required Spec Kit Version

=0.10.0

Required Tools (optional)

speckit_version: >=0.10.0

Number of Commands

4

Number of Hooks (optional)

No response

Tags

mcp, code-intelligence, context-compression, tokens, claude, spec-driven-development

Key Features

• Graph-based code navigation — replaces grep/file reads with knowledge graph queries (99% fewer retrieval tokens)
• Automatic context compression via headroom-ai (60–95% fewer prompt tokens)
• Compressed Spec Kit artifact loading — spec/plan/tasks read through headroom before reaching the LLM
• Session token savings tracking with cf_stats
• Works with Claude Code, Cursor, and Codex

Testing Checklist

• Extension installs successfully via download URL
• All commands execute without errors
• Documentation is complete and accurate
• No security vulnerabilities identified
• Tested on at least one real project

Submission Requirements

• Valid extension.yml manifest included
• README.md with installation and usage instructions
• LICENSE file included
• GitHub release created with version tag
• All command files exist and are properly formatted
• Extension ID follows naming conventions (lowercase-with-hyphens)

Testing Details

Tested on Windows 11, Python 3.14, Claude Code.
Commands validated manually. Real project test (Domika — React + Supabase)
in progress and will be documented in the repository.

Example Usage

specify extension install contextforge-mcp
cbm_index_repository(repo_path=".")
cbm_search_graph(name_pattern=".*Payment.*")
cf_stats()

Proposed Catalog Entry

"contextforge-mcp": {
  "name": "ContextForge MCP",
  "id": "contextforge-mcp",
  "version": "0.1.0",
  "description": "Integrates codebase-memory-mcp + headroom into Spec Kit — graph-based code intelligence and context compression for the implement phase.",
  "author": "capatinore",
  "license": "MIT",
  "category": "code",
  "effect": "read-write",
  "repository": "https://github.com/capatinore/contextforge-mcp",
  "download_url": "https://github.com/capatinore/contextforge-mcp/releases/download/ext-v0.1.0/contextforge-mcp-speckit-extension.zip",
  "requires": {
    "speckit_version": ">=0.10.0"
  }
}

Additional Context

No response

[capatinore]

Real-world benchmark results (Domika — React + TypeScript + Supabase, 3,048 nodes)

Query | Original tokens | Compressed tokens | Savings -- | -- | -- | -- get_architecture | 3,122 | 2,822 | 9.6% search_graph (266 functions) | 14,381 | 2,295 | 84% Session total | 17,503 | 5,117 | 70.8%

Estimated cost saved: $0.0372 (at $3/M tokens, Claude Sonnet pricing)

Notes:

• Compressor is pure-Python, cross-platform — no ML models required, works on Windows
• Automatically removes CBM fingerprint fields (fp, sp, bt) which have no informational value
• Arrays deduplicated and truncated to tool-specific limits (30 for search_graph, 60 for get_architecture)
• Token counting uses character-based estimation (len // 4) accurate for JSON + prose

Latest release: v0.2.4 — pip install contextforge-mcp