TokenLens MCP
Analyze AI conversations to estimate token costs, detect waste, and generate optimized prompts directly from your editor.
README
TokenLens MCP
Token intelligence layer for AI editors — analyze conversations, estimate costs, detect waste, and generate optimized prompts.
Quick Start
npm install
cp .env.example .env # fill in your values
npm run build
npm start # HTTP server on port 3000
npm run stdio # stdio transport for Claude Desktop
MCP Tools
| Tool | Description |
|---|---|
analyze_conversation |
Token count, cost estimate, waste detection |
improve_prompt |
Rewrite vague prompts into precise ones (local, no API) |
smart_compress |
Compress context by importance score |
estimate_cost |
Pre-flight cost estimate across all models |
get_budget_status |
Daily token budget and alert level |
weekly_summary |
Weekly usage vs last week |
export_report |
Markdown report saved to ~/.tokenlens/reports/ |
REST Endpoints (Chrome Extension)
POST /analyze-direct Analyze a conversation (no auth)
POST /improve-prompt Local prompt improvement (no auth)
POST /improve-prompt-ai AI-powered prompt improvement (user API key)
POST /generate-prompt AI prompt generation (user API key)
GET /dashboard Web UI
GET /health Health check
Configuration
Copy .env.example to .env and set:
PORT=3000
ADMIN_SECRET=<strong-random-secret>
AI provider keys are not required on the server — the Chrome Extension passes the user's own key per-request. Server-side keys are only needed if you add server-managed AI features.
Security
Your API keys are never stored on our servers. All keys are stored locally on your device only.
- The Chrome Extension saves API keys to browser
localStorage— they never leave your machine - The
/generate-promptand/improve-prompt-aiendpoints receive your key only for the duration of the API call and do not log or persist it - The MCP server stores session analytics in a local SQLite database (
~/.tokenlens/sessions.db) — no data is sent to external services - Set a strong
ADMIN_SECRETin your.envbefore exposing the server to any network
Hosted Server
The production server is available at:
https://tokenlens-mcp-production.up.railway.app
| Endpoint | URL |
|---|---|
| Health | https://tokenlens-mcp-production.up.railway.app/health |
| SSE (MCP) | https://tokenlens-mcp-production.up.railway.app/sse |
| Dashboard | https://tokenlens-mcp-production.up.railway.app/dashboard |
| Chrome Extension default server URL | https://tokenlens-mcp-production.up.railway.app |
Claude Desktop Setup
Option A — Hosted (recommended): connect via SSE to the production server.
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"tokenlens": {
"url": "https://tokenlens-mcp-production.up.railway.app/sse",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}
Option B — Local stdio: run the server on your own machine.
{
"mcpServers": {
"tokenlens": {
"command": "node",
"args": ["/path/to/TokenLens MCP/dist/stdio.js"]
}
}
}
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。