Ask your AI coding agent, in natural language, to search production logs or inspect a database — safely. Two drop-in MCP servers (MySQL, SSH logs) keep safety red lines in code: read-only by default,
View on GitHub ↗Ask your AI coding agent, in natural language, to search production logs or inspect a database — safely. Works with any MCP-capable host (Codex, Claude Code, OpenCode, Cursor, ...).
Core idea: keep the safety red lines in code, not in a prompt. Instead of trusting the model to obey instructions, the limits are enforced by validation and covered by unit tests.
Package Description Install --------- safe-mysql-mcp Read-only-by-default MySQL MCP server with SQL guardrails pip install -e ./safe-mysql-mcp ssh-logs-mcp Read-only remote log inspection MCP server with a code-enforced command allowlist pip install -e ./ssh-logs-mcp
The two packages are independent and can also be published as separate repositories.
Each package's core is an MCP server; it also ships a skill/ layer that is recommended to install on skill-capable hosts (Codex / Claude Code), so the model knows when and how to call the tools.
Host integration snippets (Codex / Claude Code / OpenCode / Cursor) are in each package README.
Letting an agent connect straight to production databases or run remote shell commands is risky. Most existing MCP servers only handle "connecting" and leave safety to the model's prompt. This project encodes the limits:
- Read-only by default, DDL / dangerous statements blocked, LIMIT injected automatically, writes must carry a WHERE clause; - Command allowlist, no ; / && / redirects / backticks / $(), no path traversal; - Production disabled by default and must be enabled explicitly; - Credentials only via environment variables or local config files, never committed.
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/wangke-112--agent-safe-tools.html)
Keep the signal. Cut the noise. Local CLI + MCP — fold verbose tool output for Claude Code, Codex, Cursor & Desktop.
dontpayfull/AInotateAnnotated screenshots for AI agents: Skitch-style arrows, numbered steps, boxes, labels, solid redaction
MohammadHijjawi97/since-cutoffFind which APIs of your pinned Python dependencies changed after your coding model's training cutoff, and give the agent short AGENTS.md / CLAUDE.md notes from a static API diff. No model calls, no AP
TeodorMCP/universal-connector-mcpUniversal MCP server: connect any OpenAPI/Swagger, GraphQL, gRPC or SOAP API to AI agents. Security-first, local, token-efficient. Any API. One server.
sainitish1609/mcp-guard🛡️ Ultra-fast local security firewall, secret sanitizer, and context token compressor for Claude Code, Cursor, and MCP AI agents.
2akouwu/reverifyStop your AI from making things up — it proposes, deterministic tools decide, every claim checked against ground truth with evidence. Grounded facts and context survive resets. Reverse engineering is
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.