RyanAlberts/best-of-Agent-Harnesses
🏆 Curated, ranked list of AI agent harnesses (100+) — plus an MCP server, llms.txt & JSON so agents can recommend them too. Rescored weekly.
Critical Issues
Security ScoreF
Vulnerabilities Found
Quality ScoreD
Maintenance55
Popularity35
Documentation26
Available Documentation
Compatibility72
Features
Agentic MCP Risk Intelligence
MCPSafe checks risks unique to AI-connected MCP servers — prompt surfaces, model-controlled tools, credentials, filesystems, and network reachability.
The current scan did not find prompt-injection, agent command, exfiltration, unbounded filesystem, or MCP SSRF patterns.
Tool metadata can steer or hijack AI clients before a user invokes the tool.
Model-selected tool arguments can reach shell/process execution.
Tool context combines credential access with outbound network capability.
AI-controlled paths may access files outside the intended workspace.
AI-controlled URLs may reach local, metadata, or internal network services.
Vulnerabilities(15)
README
No README available for this server.
View repositoryEmbed Security Badge
Add this badge to your README or documentation
[](https://mcpsafe.org/registry/ryanalberts-best-of-agent-harnesses)Server Information
- Source
- GITHUB
- Language
- python
- License
- CC-BY-SA-4.0
- Added
- Jul 20, 2026
- Updated
- Jul 20, 2026