- Conduid
- Marketplace
- #llm-security
MCP servers tagged llm-security
557 live MCP servers tagged "llm-security", ranked by trust score. Tags come from package metadata and repository topics, so this list covers servers across every category that work with llm-security.
Divergence
Precision-first security analyzer for MCP servers and agent skills. 0% false positives on traps, 96% recall.
Codelytra
Open-source, local-first AI security scanner for JavaScript and TypeScript LLM, agent, RAG, tool-calling, and MCP applications.
Onramp
Minimal MCP-over-stdio server skeleton: tool registry, var/ runtime layout, env-sync, container-ready
Nuvrail
Approval gateway for AI agents' email: an IMAP/SMTP proxy that stages every write (move, delete, send) for human sign-off. Self-hosted, AGPL-3.0.
Simdiff
The effect layer for AI-agent firewalls — simulate an action, get a canonical effect delta. Zero-dep, deterministic, fail-closed. 5 domains (fs, sql, shell, http, solana).
Peakstar Mcpguard
MCP / Agent security scanner: scan for security holes before your AI agent touches company systems. Zero dependencies, static, runs locally, sends nothing out.
Proven Gate
Deterministic provenance policy layer between LLM agents and their tools (including MCP) — blocks tool calls driven by untrusted data. No LLM, no network, fail-closed.
State Of MCP Security
Reproducible security audit of MCP servers — 75% of scanned servers exposed a medium+ hardening issue to AI agents. Round 1: official reference + popular hosted servers.
Agentomy Agent
Free-forever governance peer for AI agents. Score, flag, and audit any agent session in 30 seconds, no infrastructure. Open (MIT) system prompt plus TypeScript/Python SDK. Part of…
Toolwall
Information-flow firewall for MCP tool calls: tracks what a session has read and blocks the call that would send it out.
Agent Security Studio
Flagship AI agent security studio with risk simulation, tool approvals, RAG evals, traces, GitHub audits, tests, CI, and Pages deploy.
Aitdp
AI Threat Detection Protocol — open spec, community ruleset and reference SDK for detecting prompt injection, jailbreaks, data exfiltration and tool abuse in LLM apps and AI agent…
Razorpay Hackathon Agentic Gateway
Zero-trust payment gateway for AI agents: Ed25519-signed AP2-style mandates, two-phase reserve/settle on Razorpay, layered prompt-injection guards (regex + Llama Prompt Guard 2),…
Agentsafe Guard
Runtime governance for AI agents. Wrap a tool in three lines; every call is signed and decided outside your process. Blocked calls never reach your code.
Unmask
Reveal the hidden Unicode an LLM reads: zero-width chars, bidi overrides, invisible tag payloads, and homoglyphs in any prompt or skill file. Client-side.
Agent Boundary
A deterministic tool-call broker that bounds what an LLM agent can actually do. Indirect prompt injection is assumed to succeed; authorisation lives outside the model.
Surface Scan
Read an MCP client config and report what each server can actually reach — before you mount it.
Samar MCP
🛡️ Ephemeral privacy & security middleware for Model Context Protocol (MCP). Zero-persistence credential masking & fail-closed execution guard for AI agents.
Valvur
Offline security scanning for AI-generated code: seven open-source scanners and AI-specific checks in one command. It finds secrets, vulnerable, hallucinated and malicious depende…
Agent Cve Scanners
Free offline scanners for 53 CVEs in 9 AI-agent frameworks + MCP tool pinning. No telemetry.
RogueAgent Testbed
Georgia Tech master's in cybersecurity research on cascading indirect prompt injection attacks. Author: Fahid Ahmed
Sleeper
Local-first agent tool-call assurance lab: injection → tool action → control gate → evidence.
Claude MCP Allow
Generate least-privilege Claude Code permission rules for MCP tools from their live annotations: readOnlyHint tools go to allow, everything else to ask, one mcp__server__tool rule…
Agentic Semgrep Rules
Semgrep rules for AI agent code: static analysis for LLM applications in Python, TypeScript and JavaScript. Finds model output reaching exec, shells, SQL, URLs, files and HTML, us…
Guardian
MCP Guardian acts as a proxy service for remote MCP endpoints, and constantly polls them to make sure they haven't been compromised or modified.
Glyph
🔮 MCP security scanner — find tool poisoning, credential leaks, and insecure transports in AI agent configurations
Guard
Security scanner for MCP (Model Context Protocol) servers: detects tool poisoning, prompt injection, command injection, path traversal and leaked secrets. SARIF, GitHub Action, si…
AI Surface
Find and govern AI attack surfaces in application code, at PR time and inside your AI coding tool (MCP server, Claude Code hook). Free, OSS, runs offline.
Www Project Agent Memory Guard
OWASP Foundation web repository
AI Surface
Find and govern AI attack surfaces in application code, at PR time and inside your AI coding tool (MCP server, Claude Code hook). Free, OSS, runs offline.
Template
Secure MCP server template in TypeScript and Python: stdio by default, optional streamable HTTP behind bearer auth, rate limit and body cap, bound to 127.0.0.1; bounded file and n…
Www Project Agent Memory Guard
OWASP Foundation web repository
Firewall
Scan and gate Model Context Protocol servers for tool poisoning, rug-pulls, and indirect prompt injection.
Mcpguard
Local firewall, scanner, and audit logger for MCP servers. Auto-discovers Claude Desktop / Cursor / Windsurf / VS Code configs and flags risky tool servers — leaked secrets, curl|…
LLM Blue Team Guardrail
A layered LLM guardrail (fast heuristics + an LLM judge) against prompt injection, with a guarded RAG pipeline that quarantines poisoned documents and an MCP server for AI agents.…
Audit
Python security scanner for Model Context Protocol (MCP) servers — find prompt injection, over-broad permissions, weak input validation, and credential leaks before your AI agent…
MCPGuard
MCPGuard is the open-source CI gate for MCP servers — it tests protocol compliance, scans for security vulnerabilities, validates tool behavior, and benchmarks performance before…
Readonly SQL MCP
An MCP server that gives an AI agent read-only SQL access: three independent locks, each proven by breaking it on purpose, plus agent evaluations.
Policy Gateway
Runtime policy enforcement and adversarial evaluation for MCP tool calls
Red Team
mcp security scanner. mid-2025. deprecated.
Warden
A security proxy for the Model Context Protocol. Every tool call gets integrity-checked, policy-checked, inspected, and audited before it runs.
Agenticqa Scan Action
Analyze AI codebases to identify integration points, security risks, and test coverage gaps in one step with comprehensive reports.
SkillScanner
Universal AI agent skill security scanner. 5-layer deep defense for prompt injection, MCP tool poisoning, encoding obfuscation. Zero dependencies. Bilingual EN/ZH.
Agentpassport
Verifiable identity, scoped delegation, revocation, and signed action receipts for AI agents.
Agent Cordon
Zero-dependency prompt-injection and data-exfiltration firewall for LLM agents: scans untrusted text and outbound actions, sees through obfuscation, guards the MCP/tool boundary.
Scan
CLI security scanner for MCP (Model Context Protocol) servers — detects ANSI escape injection, over-permissive tools, prompt injection vectors, and insecure transport
Ares
Offensive security framework for autonomous AI agents — recon, foothold, provenance desync, coherence-evasion pacing, recovery poisoning, and exfil, gated by signed attestations a…
Agent Vault
A local credential proxy for coding agents: the agent sends a placeholder and never holds the secret. Free and open source from ioLite Labs.
Agentx Security SDK
Runtime action firewall for AI agents: the keyless Shield. Screens catastrophic tool calls (DROP TABLE, SSRF, secret exfil) in-process, no key. Watches by default; set the posture…
Mayi
Argument-level authority for AI agent tool calls. Every argument has one owner (app, model, human or derived). Plain deterministic rules, no judging model. Held calls wait for a p…