signalforge
Local-first MCP server that analyzes and compiles prompts for multiple LLM clients, detecting intent, missing context, and waste risks to generate cost-aware, task-specific prompts.
README
SignalForge
SignalForge is a local-first MCP prompt intent compiler. It turns vague user requests into task-specific, cost-aware prompts for Codex first, while keeping adapters for Claude Code, Cursor, ChatGPT, Kimi, and generic LLM workflows.
It is not another prompt prettifier. SignalForge detects intent, missing context, workflow shape, context waste, and likely failure modes before you spend a full model run.
What It Does
- Analyzes rough prompts for task type, clarity, missing context, and waste risks.
- Compiles prompts for Codex, Claude Code, Cursor, ChatGPT, Kimi, or generic clients.
- Suggests split plans for large tasks instead of one expensive mega-prompt.
- Provides reusable prompt patterns for common workflows.
- Saves only user-approved patterns locally in
.signalforge/patterns.json. - Does not save raw prompts by default.
Install
npm install
npm run build
Run As An MCP Server
Codex:
codex mcp add signalforge -- node C:\Users\Karthikey G\OneDrive\Documents\signalforge\dist\stdio.js
Claude Desktop / Claude Code style config:
{
"mcpServers": {
"signalforge": {
"command": "node",
"args": ["C:\\Users\\Karthikey G\\OneDrive\\Documents\\signalforge\\dist\\stdio.js"]
}
}
}
MCP Inspector:
npm run inspect
CLI Smoke Test
npm run smoke
Or run directly:
node dist/cli.js analyze "build me a portfolio app" --target codex
node dist/cli.js compile "build me a portfolio app" --target codex
node dist/cli.js compile "compare these two startup ideas" --target kimi
node dist/cli.js pattern feature_build --target codex
Use With Kimi
Kimi does not need MCP support for this flow. Compile the prompt locally, then paste the output into Kimi:
node dist/cli.js compile "help me design an AI automation product" --target kimi
MCP Tools
analyze_prompt: intent, clarity, missing context, waste risks, workflow.compile_prompt: ready-to-use prompt with assumptions, split plan, and cost-saving notes.get_prompt_pattern: reusable template for a use case.estimate_prompt_waste: likely failure mode and cheaper prompting route.save_pattern: store a user-approved pattern locally.list_patterns: list saved local patterns.
MCP Resources
signalforge://patternssignalforge://adapters/codexsignalforge://adapters/claude-codesignalforge://adapters/cursorsignalforge://adapters/chatgptsignalforge://adapters/kimisignalforge://adapters/genericsignalforge://history/recent
Development
npm run test
npm run build
npm run smoke
The compiler is deterministic in v1. Optional LLM refinement can be added later behind an environment flag, but the core should remain useful without any API key.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。