OWASP Agentic MCP
You can run read-only, idempotent OWASP Agentic AI security assessments and checks via this MCP server.
Perform a full OWASP Agentic AI Top 10 security assessment for an agent using its name and security-control flags.
Detect prompt injection attack patterns in input text.
Check tool definitions for tool poisoning risk from name/description manipulation and trust signals.
Assess excessive agency by comparing available/used tools and permissions against least-privilege controls.
Assess cross-context data exposure and leakage risks, including isolation, PII detection, logging, and third-party sharing.
Use basic usage without authentication, with optional API key for Pro/Enterprise tiers; free tier limited to 10 calls/day.
Receive in-memory analysis output only; tools are stateless, have no side effects, and are not a substitute for human review in production decisions.
Security assessment against the OWASP Top 10 for Agentic AI, providing tools to detect prompt injection, tool poisoning, excessive agency, and data leakage.
Owasp Agentic MCP
OWASP Top 10 for Agentic AI security MCP server — prompt injection detection, tool poisoning, exc...
OWASP Top 10 for Agentic AI security MCP server — prompt injection detection, tool poisoning, excessive agency, data leakage assessment
OWASP Agentic MCP
OWASP Top 10 for AI Agents Security Assessment — Prompt Injection, Tool Poisoning, Data Leakage
🚀 Quick Start
# Install via pip
pip install owasp_agentic_mcp
# Or install via Smithery
npx -y @smithery/cli@latest install owasp-agentic-mcp --client claudeRelated MCP server: Agent Delegation MCP
✨ Features
MCP protocol compliant
Easy installation
Well-documented API
Production-ready
Active maintenance
📖 Documentation
🛡️ Compliance
This MCP server is built with EU AI Act compliance built-in:
✅ Article 9 — Risk Management System
✅ Article 13 — Transparency & Instructions for Use
✅ Article 15 — Bias Detection & Testing
✅ Article 26 — FRIA Support (where applicable)
✅ Article 50 — AI Content Watermarking (where applicable)
Need help getting compliant? Book a free 15-min diagnostic →
How MEOK charges
MEOK has no subscriptions and no tiers. You pay per job, and only when the agreed artifact has been delivered to you. There is no public price list. See https://meok.ai/pricing.
This repository's source is MIT-licensed and free to self-host.
🤝 Part of the MEOK Ecosystem
This server is part of the MEOK AI Labs ecosystem — 300+ MCP servers for sovereign AI governance.
Domain | Purpose |
EU AI Act compliance marketplace | |
AI safety & monitoring | |
Sovereign AI platform | |
Legacy modernization |
📜 License
MIT © CSOAI-ORG
Tools
Tool | Description | Parameters |
| Full OWASP Agentic Top 10 security scan |
|
| Assess against specific OWASP Agentic categories |
|
| Test for prompt injection vulnerabilities |
|
| Check for tool poisoning risks |
|
| Assess agency level vs minimum required |
|
Installation
pip install mcpClaude Desktop / Cursor / VS Code / Windsurf
{
"mcpServers": {
"owasp-agentic": {
"command": "python",
"args": ["path/to/server.py"]
}
}
}Usage Examples
Full agent security scan
{
"agent_config": {
"system_prompt": "You are a helpful assistant that can access email, calendar, and file system.",
"allowed_tools": ["send_email", "read_calendar", "write_file"]
},
"permissions": ["email:send", "calendar:read", "files:write"]
}Check prompt injection
{
"system_prompt": "You are a financial advisor bot",
"user_input_template": "{user_query} Please provide investment advice."
}Available Tools
5 toolsassess_agent_securityA
Full OWASP Agentic AI Top 10 security assessment.
Behavior: This tool is read-only and stateless — it produces analysis output without modifying any external systems, databases, or files. Safe to call repeatedly with identical inputs (idempotent). Free tier: 10/day rate limit. Pro tier: unlimited. No authentication required for basic usage.
When to use: Use this tool when you need structured analysis or classification of inputs against established frameworks or standards.
When NOT to use: Not suitable for real-time production decision-making without human review of results.
Args: agent_name (str): The agent name to analyze or process. has_input_validation (bool): The has input validation to analyze or process. has_output_filtering (bool): The has output filtering to analyze or process. has_tool_allowlist (bool): The has tool allowlist to analyze or process. has_least_privilege (bool): The has least privilege to analyze or process. has_context_isolation (bool): The has context isolation to analyze or process. has_action_logging (bool): The has action logging to analyze or process. has_auth_between_agents (bool): The has auth between agents to analyze or process. has_resource_limits (bool): The has resource limits to analyze or process. has_dependency_scanning (bool): The has dependency scanning to analyze or process. has_alignment_testing (bool): The has alignment testing to analyze or process. api_key (str): The api key to analyze or process.
Behavioral Transparency: - Side Effects: This tool is read-only and produces no side effects. It does not modify any external state, databases, or files. All output is computed in-memory and returned directly to the caller. - Authentication: No authentication required for basic usage. Pro/Enterprise tiers require a valid MEOK API key passed via the MEOK_API_KEY environment variable. - Rate Limits: Free tier: 10 calls/day. Pro tier: unlimited. Rate limit headers are included in responses (X-RateLimit-Remaining, X-RateLimit-Reset). - Error Handling: Returns structured error objects with 'error' key on failure. Never raises unhandled exceptions. Invalid inputs return descriptive validation errors. - Idempotency: Fully idempotent — calling with the same inputs always produces the same output. Safe to retry on timeout or transient failure. - Data Privacy: No input data is stored, logged, or transmitted to external services. All processing happens locally within the MCP server process.
Inputs: agent_name (required), has_input_validation, has_output_filtering, has_tool_allowlist, +9 more. Returns: result
check_data_leakageA
Assess cross-context data exposure risks.
Behavior: This tool is read-only and stateless — it produces analysis output without modifying any external systems, databases, or files. Safe to call repeatedly with identical inputs (idempotent). Free tier: 10/day rate limit. Pro tier: unlimited. No authentication required for basic usage.
When to use: Use this tool when you need structured analysis or classification of inputs against established frameworks or standards.
When NOT to use: Not suitable for real-time production decision-making without human review of results.
Args: agent_name (str): The agent name to analyze or process. has_context_isolation (bool): The has context isolation to analyze or process. has_session_boundaries (bool): The has session boundaries to analyze or process. has_pii_detection (bool): The has pii detection to analyze or process. has_output_sanitization (bool): The has output sanitization to analyze or process. shares_memory_across_users (bool): The shares memory across users to analyze or process. logs_contain_user_data (bool): The logs contain user data to analyze or process. third_party_data_sharing (bool): The third party data sharing to analyze or process. api_key (str): The api key to analyze or process.
Behavioral Transparency: - Side Effects: This tool is read-only and produces no side effects. It does not modify any external state, databases, or files. All output is computed in-memory and returned directly to the caller. - Authentication: No authentication required for basic usage. Pro/Enterprise tiers require a valid MEOK API key passed via the MEOK_API_KEY environment variable. - Rate Limits: Free tier: 10 calls/day. Pro tier: unlimited. Rate limit headers are included in responses (X-RateLimit-Remaining, X-RateLimit-Reset). - Error Handling: Returns structured error objects with 'error' key on failure. Never raises unhandled exceptions. Invalid inputs return descriptive validation errors. - Idempotency: Fully idempotent — calling with the same inputs always produces the same output. Safe to retry on timeout or transient failure. - Data Privacy: No input data is stored, logged, or transmitted to external services. All processing happens locally within the MCP server process.
Inputs: agent_name (required), has_context_isolation, has_session_boundaries, has_pii_detection, +6 more. Returns: result
check_excessive_agencyA
Assess agent for excessive permissions (least privilege).
Behavior: This tool is read-only and stateless — it produces analysis output without modifying any external systems, databases, or files. Safe to call repeatedly with identical inputs (idempotent). Free tier: 10/day rate limit. Pro tier: unlimited. No authentication required for basic usage.
When to use: Use this tool when you need structured analysis or classification of inputs against established frameworks or standards.
When NOT to use: Not suitable for real-time production decision-making without human review of results.
Args: agent_name (str): The agent name to analyze or process. tools_available (int): The tools available to analyze or process. tools_used_in_task (int): The tools used in task to analyze or process. has_approval_gates (bool): The has approval gates to analyze or process. has_scope_limits (bool): The has scope limits to analyze or process. can_access_filesystem (bool): The can access filesystem to analyze or process. can_access_network (bool): The can access network to analyze or process. can_execute_code (bool): The can execute code to analyze or process. can_modify_data (bool): The can modify data to analyze or process. can_send_communications (bool): The can send communications to analyze or process. api_key (str): The api key to analyze or process.
Behavioral Transparency: - Side Effects: This tool is read-only and produces no side effects. It does not modify any external state, databases, or files. All output is computed in-memory and returned directly to the caller. - Authentication: No authentication required for basic usage. Pro/Enterprise tiers require a valid MEOK API key passed via the MEOK_API_KEY environment variable. - Rate Limits: Free tier: 10 calls/day. Pro tier: unlimited. Rate limit headers are included in responses (X-RateLimit-Remaining, X-RateLimit-Reset). - Error Handling: Returns structured error objects with 'error' key on failure. Never raises unhandled exceptions. Invalid inputs return descriptive validation errors. - Idempotency: Fully idempotent — calling with the same inputs always produces the same output. Safe to retry on timeout or transient failure. - Data Privacy: No input data is stored, logged, or transmitted to external services. All processing happens locally within the MCP server process.
Inputs: agent_name (required), tools_available, tools_used_in_task, has_approval_gates, +8 more. Returns: result
check_prompt_injectionA
Check text for prompt injection attack patterns.
Behavior: This tool is read-only and stateless — it produces analysis output without modifying any external systems, databases, or files. Safe to call repeatedly with identical inputs (idempotent). Free tier: 10/day rate limit. Pro tier: unlimited. No authentication required for basic usage.
When to use: Use this tool when you need structured analysis or classification of inputs against established frameworks or standards.
When NOT to use: Not suitable for real-time production decision-making without human review of results.
Args: input_text (str): The input text to analyze or process. api_key (str): The api key to analyze or process.
Behavioral Transparency: - Side Effects: This tool is read-only and produces no side effects. It does not modify any external state, databases, or files. All output is computed in-memory and returned directly to the caller. - Authentication: No authentication required for basic usage. Pro/Enterprise tiers require a valid MEOK API key passed via the MEOK_API_KEY environment variable. - Rate Limits: Free tier: 10 calls/day. Pro tier: unlimited. Rate limit headers are included in responses (X-RateLimit-Remaining, X-RateLimit-Reset). - Error Handling: Returns structured error objects with 'error' key on failure. Never raises unhandled exceptions. Invalid inputs return descriptive validation errors. - Idempotency: Fully idempotent — calling with the same inputs always produces the same output. Safe to retry on timeout or transient failure. - Data Privacy: No input data is stored, logged, or transmitted to external services. All processing happens locally within the MCP server process.
Inputs: input_text (required), caller, api_key. Returns: result
check_tool_poisoningA
Check a tool for name/description manipulation (tool poisoning).
Behavior: This tool is read-only and stateless — it produces analysis output without modifying any external systems, databases, or files. Safe to call repeatedly with identical inputs (idempotent). Free tier: 10/day rate limit. Pro tier: unlimited. No authentication required for basic usage.
When to use: Use this tool when you need structured analysis or classification of inputs against established frameworks or standards.
When NOT to use: Not suitable for real-time production decision-making without human review of results.
Args: tool_name (str): The tool name to analyze or process. tool_description (str): The tool description to analyze or process. tool_source (str): The tool source to analyze or process. has_signature_verification (bool): The has signature verification to analyze or process. has_description_hash (bool): The has description hash to analyze or process. from_trusted_registry (bool): The from trusted registry to analyze or process. api_key (str): The api key to analyze or process.
Behavioral Transparency: - Side Effects: This tool is read-only and produces no side effects. It does not modify any external state, databases, or files. All output is computed in-memory and returned directly to the caller. - Authentication: No authentication required for basic usage. Pro/Enterprise tiers require a valid MEOK API key passed via the MEOK_API_KEY environment variable. - Rate Limits: Free tier: 10 calls/day. Pro tier: unlimited. Rate limit headers are included in responses (X-RateLimit-Remaining, X-RateLimit-Reset). - Error Handling: Returns structured error objects with 'error' key on failure. Never raises unhandled exceptions. Invalid inputs return descriptive validation errors. - Idempotency: Fully idempotent — calling with the same inputs always produces the same output. Safe to retry on timeout or transient failure. - Data Privacy: No input data is stored, logged, or transmitted to external services. All processing happens locally within the MCP server process.
Inputs: tool_name (required), tool_description (required), tool_source, has_signature_verification, +4 more. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Each tool targets a distinct security aspect (overall assessment, data leakage, excessive agency, prompt injection, tool poisoning) with clear, non-overlapping descriptions, ensuring an agent can easily distinguish them.
All tool names follow a consistent verb_noun pattern (e.g., check_data_leakage, assess_agent_security) with no mixed conventions or deviations, making navigation predictable.
Five tools is a well-scoped count for a security assessment server, covering key OWASP agentic AI vulnerabilities without being too sparse or overwhelming.
The set covers major OWASP Top 10 areas but lacks tools for supply chain security or model theft. However, the comprehensive assess_agent_security tool partially fills gaps, making the surface reasonably complete.
Maintenance
Related MCP Connectors
AI-security knowledge as MCP: standards-mapped tools (OWASP, NIST, MITRE) for AI agents.
MEOK MCP Hardening MCP — automated security red-team for any MCP server. Maps OWASP LLM Top 10
Agent Orchestrator MCP Server by MEOK AI Labs
LLM Orchestration MCP Agent
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceAn MCP server that exposes over 20 standard penetration testing utilities, such as Nmap, SQLMap, and OWASP ZAP, as callable tools for AI agents. It enables natural language control over complex security workflows for automated and interactive penetration testing.120-
- AlicenseNot gradedqualityCmaintenanceAgent Delegation - MCP server providing AI-powered tools and automation by MEOK AI Labs6 npm148 PyPIMIT
- AlicenseNot gradedqualityAmaintenanceAn MCP server that exposes a 60+ tool security and threat-intel stack to AI agents, enabling secret scanning, Sigma rule generation, ransomware lookup, OSINT, and deep research.1MIT
- AlicenseNot gradedqualityDmaintenanceAI-Powered Red Team MCP Server enabling autonomous penetration testing via Model Context Protocol with 44+ security tools for AI agents.17MIT