MCP servers that ship as code — a repository or a package you install and run locally (usually over stdio). They have no hosted endpoint, so we cannot probe or grade them: instead each one shows its repository, install command, license, stars and when it was last updated.
Collected from GitHub repositories and npm, PyPI, Cargo, Go, Maven, NuGet and RubyGems packages tagged as MCP servers, the official MCP Registry, Docker's MCP catalog, ModelScope, Baidu MCP World, Cline and other public directories; one entry per repository. A server that also offers a hosted endpoint is listed under Remote instead. These entries are not counted in our reachability, verification or readiness statistics.
-
mcp-sentinelgithub.com/jakeefr/mcp-sentinelupdated this year
Runtime security scanner for MCP tool poisoning. Native Claude Code plugin.
MIT ★ 3 -
Proof of impact for AI attack surfaces. Probe MCP servers, catch tool poisoning, ship bounty-grade evidence.
MIT ★ 1 -
MCP server security auditor — tool poisoning, ASCII smuggling detection, prompt injection, OWASP Agentic Top 10. First OSS tool for MCP security.
AGPL-3.0 ★ 2 -
MCP Server Secret Injector (envFile / AES-256-GCM Vault), Tool-Poisoning Scanner & Real-Time stdio JSON-RPC 2.0 Firewall.
MIT ★ 1 -
mcp-schema-sentinelgithub.com/amurlaniakea/mcp-schema-sentinelupdated this year
Read-only sensor that detects MCP tool-poisoning / rug-pull attacks: alerts when a trusted tool's schema or description changes silently.
AGPL-3.0 ★ 1 -
Reproducible MCP tool-poisoning (line-jumping) demo, contained by an independent plan-vs-authorize gate. Own testbed, mock server, fake secret — attacks nobody.
MIT -
MCP servers for offensive tooling (Mythic C2, BloodHound, nmap) plus AI attack reference: prompt injection, insecure output handling, function-calling abuse, data exfiltration, data poisoning. Authorized use only.
MIT ★ 3 -
Prompt-injection scanner for AI agents: deterministic, documented rules (instruction override, jailbreaks, exfiltration URLs, tool hijack/MCP tool poisoning, hidden Unicode, encoded payloads) returning a risk score, categories and spans. Library + x402 pa
npm MIT -
MCP server that lets LLM agents drive a data-poisoning detection research pipeline: 28 typed tools, plus a mock backend to run it end to end.
MIT -
MCP injection / prompt-poisoning / SSRF scanner. 30+ canonical rules covering the April 2026 Anthropic MCP RCE class, mcp-server-git CVE chain, DockerDash injection chain, OWASP LLM Top 10 + GenAI Red Team v1. Scans live MCP /tools/list endpoints OR pasted JSON. Free 5/day, Pro £29/mo for unlimited + signed safety reports auditors and procurement accept. By MEOK AI Labs.
pypi MIT License -
@mseep/ipinfostale

npm UNLICENSED -
poisonvectorsgithub.com/jamesburton/poison-vectorsupdated this year
MCP server for semantic vector search with relevance feedback (poison-vector technique). Exposes embed, rerank, and feedback tools over stdio.
nuget -
MCP SafeTripwire · MCP 安全绊线。Windows MCP security broker: tool-poisoning checks, manifest integrity, behavior baselines, blacklists, blocking and audit logs. By luke0485. Preview; macOS/Linux planned.
MIT ★ 1 -
MCP server that scores text, web pages and MCP tool definitions for prompt injection and tool poisoning. Free tier, then pay per call via x402.
MIT -
🔌 Exposes OWASP Top 10 for Agentic AI security over MCP — prompt injection detection, tool poisoning, excessive agency, data leakage assessment
pypi MIT -
asoftimegithub.com/amatayomosley-web/asofupdated this year
Temporal-awareness hook for tool-using LLMs. Surfaces freshness verdicts on in-context data so stale information doesn't poison fresh reasoning. Adapters for Claude Code, Antigravity, generic harnesses (LangGraph, CrewAI, raw SDK pipelines), and an MCP server for any MCP-compatible client.
pypi MIT -
Security scanner for MCP servers, as a CLI, GitHub Action and MCP server: tool poisoning, prompt injection, secret leaks and excessive permissions, with SARIF output for CI.
npm Apache-2.0 -
MCP security scanner (CLI + MCP server): detects tool poisoning, prompt injection, and rug-pulls in MCP manifests. Also an OWASP LLM Top 10 prompt-injection/jailbreak CLI for LLM apps.
npm MIT ★ 3 -
cuba-memorysgithub.com/leandropg19/cuba-memorysupdated this year
Persistent memory MCP server for AI agents (v0.18). 29 tools, self-growing graph without MCP sampling, poisoning quarantine: bitemporal facts, sqlx-migrate, BM25+MMR+OOD hybrid retrieval, graph metrics, conformal prediction, Hebbian/BCM, project scoping, tamper-evident audit chain (21 CFR Part 11-oriented), tree-sitter codegraph, git-native sync hooks. Rust + PostgreSQL + pgvector.
pypi Apache-2.0 -
MCPの道具に仕込まれた指示と、承認後の書き換えを見つける(公式の参照サーバーで検証・GitHub Action対応)
MIT -
agentauditgithub.com/agentaudit-dev/agentaudit-cliupdated this year
Security scanner for AI agent packages — CLI + MCP server
npm AGPL-3.0 -
Small Print's open tools: the disclosure lab, and soon the command, the MCP server and the plugin
★ 2 -
A deliberately vulnerable MCP server built as a synthetic test fixture — not for installation or connection. Ships the audit report mcpaudit returns for it.
MIT -
agentseal-mcp-intelgithub.com/agentseal/agentseal-mcp-intelupdated this year
MCP server for AgentSeal — find safe MCP servers, check security before installing, and scan your AI dev environment from inside your AI coding assistant
npm FSL-1.1-Apache-2.0 -
AegisGate MCP — Standalone Model Context Protocol security server with 23 security layers, vendored ONNX runtime, neural threat detection (L3). SSE streaming. Session management. Zero module dependencies. Multi-arch (amd64/arm64). Apache 2.0.
oci Apache-2.0 -
55-tool MCP server for security scanning of MCP servers. Runtime inspection, AST-based SAST, config audit, dependency analysis, OWASP MCP Top 10 compliance. OAuth, TLS, fuzz testing, prompt injection, tool mutation detection. 100% local, zero external API calls.
MIT ★ 5 -
MCP server for Power Strip Security: read web pages, scan content for prompt injection, and audit MCP servers and agent skills, paid per call with x402 (USDC on Base) from your own wallet, with a spending cap.
npm MIT -
MCP server for AI coding agents (Claude Code, Cursor, Cline, etc.): verify packages before installing, assess MCP servers before adding them, scan for secrets, audit agent configs, and self-review staged changes before committing.
npm MIT -
HISTOR — public transparency log of MCP tool definitions: reads what remote MCP endpoints in the official registry advertise (never calls a tool), signs it as MTL/1 labels, appends them to an RFC 9162 Merkle log, and answers /check: is what a client received what was observed? Never calls a server safe.
pypi MIT -
Anti-poisoning memory store for agents: provenance, versioning, quarantine gate
MIT ★ 1 -
A shared fix-pool for coding agents with a write-time poisoning shield — every post_solution is screened for prompt-injection and leaked secrets before it can reach a reading agent.
Apache-2.0 ★ 1 -
One local work board for AI agents from every vendor: claims that expire, review gates, verified checks, a guard against poisoned memory, dreaming and agents-only teams. MCP plugin, WORKBOARD.md and HTTP API.
MIT