janus-kinematic-drive
Gives your agent a private Kinematic Drive: a deterministic memory engine for sub-millisecond exact recall over code and docs with no hallucination. Also provides self-serve checkout and metered pay-per-call tools for AI agents.
README
<!-- mcp-name: io.github.janus-ubos-republic/janus-kinematic-drive -->
Janus — UBOS Kinematic Drive concierge (MCP server)
Talk to Janus, the UBOS concierge, right inside your coding agent — to get a private Kinematic Drive: a deterministic memory engine that gives your agent sub-millisecond, exact recall over your own private code & docs, with no LLM in the retrieval path — so it cannot hallucinate. Each drive runs isolated on its own box; your data never leaves it.
⚡ Try the live test first — no install
Open https://welcome-to-ubos.com and type a symbol (ThreadPoolExecutor, urlencode, JSONDecoder). You'll watch the real engine find the exact source in ~1ms, deterministically — that's the engine your agent gets, pointed at your own repo.
Buy & pay — for humans and agents
kinematic_checkout {email}— self-serve USDC checkout: returns an order + pay instructions (Solana wallet, memo = order id, amount).kinematic_checkout_status {order_id}returns your provision token once paid.ubos_store_products/ubos_store_call {product, api_key, payload}— the UBOS x402 metered store: governor (constitutional reality-check, $0.005/call) and hotvessel (metered LLM, $0.01/call). No account: your first call returns HTTP-402-style pay instructions (USDC with memo = your api_key); pay, wait ~15s, retry.
Machines buying from machines — no card, no signup, pure on-chain.
Why it's different
- Deterministic — your files become typed "gear teeth"; a query meshes into them over exact candidates only. Same query → same result, every time. Zero hallucination.
- Private — one isolated drive per client, on its own box. Your data never touches a cloud LLM or any shared store.
- Fast — a read-only mmap index + a tiny native scorer.
| Metric | Value |
|---|---|
| Recall latency | ~1ms p50 (~1.7ms p95) |
| Accuracy | MRR@10 0.977 · hit@10 100% |
| Hallucinations | 0 — exact meshing only |
Install
pip install ubos-janus-mcp
Claude Code
claude mcp add janus -- ubos-janus-mcp
Codex (~/.codex/config.toml)
[mcp_servers.janus]
command = "ubos-janus-mcp"
Gemini CLI (settings.json)
{ "mcpServers": { "janus": { "command": "ubos-janus-mcp" } } }
Works with Claude Code, Codex, Gemini CLI, Cline, Cursor — anything that speaks MCP.
Tools
| Tool | What it does |
|---|---|
janus_ask |
Ask Janus anything about getting/using a Kinematic Drive (grounded + live). |
kinematic_pricing |
Current price (live source of truth). |
kinematic_status |
Is a drive up and running? |
kinematic_provision |
Request a private Kinematic Drive (managed onboarding). |
kinematic_checkout |
Create a self-serve USDC order and receive Solana payment instructions. |
kinematic_checkout_status |
Poll an order until payment clears and a provision token is ready. |
ubos_store_products |
Discover the live UBOS pay-per-call agent tools and prices. |
ubos_store_call |
Call a metered tool or receive self-describing HTTP-402 payment instructions. |
How it works
This MCP server is a thin client: it proxies your agent's calls to the Janus concierge at JANUS_URL (default https://api.welcome-to-ubos.com/janus), which answers grounded on a live knowledge drive — the product, fronting the product. Your private data is never sent here; this is the storefront. Once you have a drive, you connect to it with the separate colony-gear MCP server.
Links
- Live page + test: https://welcome-to-ubos.com
- PyPI: https://pypi.org/project/ubos-janus-mcp/
- Source: https://github.com/janus-ubos-republic/janus-kinematic-drive
— UBOS Republic · constitutional infrastructure for AI agents
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。