Hoop
One gateway in front of every protocol. Same policy across MCP, LLMs, databases and containers. Wire-level enforcement at under 5ms.
One gateway in front of every protocol. Same policy across MCP, LLMs, databases and containers. Wire-level enforcement at under 5ms.
暂未识别到可直接复制的 MCP 配置,请查看 GitHub README。后台管理员可以补充配置。
NEW · User MCP server — How AI coding agents get production access safely. Read more → One Gateway Between Your Team and Your Infrastructure. hoop.dev is a layer 7 gateway that masks sensitive data, blocks dangerous commands, approves risky writes, and records every session inline, before anything reaches your infrastructure. Engineers · AI Agents · MCP Clients · Services · Support/QA Open-source. Used by NYSE-listed companies. 5,000+ databases protected. Quick start · How it works · Connectors · vs alternatives · What's new · Docs --- ## What is hoop? hoop is an open-source layer 7 gateway that sits between users (engineers, AI agents, service accounts) and infrastructure (databases, Kubernetes clusters, servers, APIs). Every query and command passes through at the wire protocol level, where the gateway can: - **Mask sensitive data in responses** — ML-powered classification, not regex pattern matching, applied before bytes leave the gateway. - **Block dangerous commands before they execute** — `DROP TABLE`, `rm -rf`, `DELETE` without `WHERE`, configurable per role and per backend. - **Require human approval for risky operations** — Slack or Teams workflow, time-bound, fully logged. - **Record every session** — full replay of SQL, shell, kubectl, and HTTP traffic, indexed by user, table, and query. No agents on endpoints. No schema discovery. No code changes. Deploy the gateway, connect your identity provider, define your rules. --- ## Who is hoop for? Teams where engineers or AI agents access production infrastructure. If your developers run queries against databases with customer PII, execute commands on production Kubernetes clusters, or use AI coding assistants against real systems, hoop gives you visibility and control over what happens inside those sessions and what data is allowed to leave them. --- ## The problem, concretely An engineer pulls recent payments to investigate a customer report: **❌ Without hoop** ```sql SELECT * FROM payments LIMIT 10; ``` ``` id | customer_email | card_number | amount | status -------+-------------------------+----------------------+--------+---------- 84021 | jane.thompson@gmail.com | 4532-1024-5678-9012 | 49.99 | settled 84022 | mreyes@acmecorp.io | 5412-7510-3344-1182 | 120.00 | settled 84023 | k.patel@protonmail.com | 4716-9923-1144-5577 | 24.99 | refunded 84024 | dlin@northwind.co | 5577-3344-9911-2266 | 89.50 | settled 84025 | tyler.s@gmail.com | 4111-2222-3333-4444 | 15.00 | failed ... ``` 10 rows of real card numbers and emails. Now in `psql` history, in the screenshot the engineer pasted into Slack, and in the CSV they exported to debug locally. **✅ With hoop** ```sql SELECT * FROM payments LIMIT 10; ``` ``` id | customer_email | card_number | amount | status -------+-----------------------+---------------------+--------+---------- 84021 | j****@*****.com | **-**-****-9012 | 49.99 | settled 84022 | m****@*******.io | **-**-****-1182 | 120.00 | settled 84023 | k****@*********.com | **-**-****-5577 | 24.99 | refunded 84024 | d****@*********.co | **-**-****-2266 | 89.50 | settled 84025 | t****@*****.com | **-**-****-4444 | 15.00 | failed ... ``` Engineers can still debug using amounts, statuses, and timestamps. PII never leaves the gateway. --- An AI agent fixing a bug at 3AM: **❌ Without hoop** > cla...
A collection of MCP servers.
Chrome DevTools for coding agents
Enhanced ChatGPT Clone: Features Agents, MCP, DeepSeek, Anthropic, AWS, OpenAI, Responses API, Azure, Groq, o1, GPT-5, M...
Playwright MCP server
GitHub's official MCP Server
🚀 The fast, Pythonic way to build MCP servers and clients.