AgentSeal

Overview

AgentSeal is an open-source security scanner and toolkit for evaluating AI agent prompts, checking MCP servers and finding vulnerabilities in agent setups. It combines adversarial scans with ongoing monitoring: users can test prompts for extraction or injection risks, inspect local agent configurations and skill files, and monitor those files and MCP configurations for changes.

The product includes a command-line interface and a web dashboard. The dashboard provides scan results, prompt management, security monitoring and GitHub integration settings. AgentSeal is freemium, with a free plan and a Pro plan whose price is not listed. Readers comparing products in this area can browse AI Red Teaming Tools.

Key features

Prompt and agent checks

The scan command runs adversarial probes against system prompts to look for extraction and instruction-injection weaknesses. The FAQ describes 311 probes covering extraction, injection, MCP tool poisoning, RAG poisoning and multimodal attacks. The free plan includes 30 basic probes; the 311-probe coverage is associated with Pro.

Guard scans agent configurations and skill files on a machine. Watch monitors those files and MCP configurations for changes, supporting continuous monitoring alongside individual scans.

MCP server scanning

Scan-MCP runs MCP servers in a sandbox and tests tool behavior using adversarial payloads. It supports OAuth for authenticated remote servers. Separately, AgentSeal's MCP registry says that each server goes through a seven-stage security pipeline before listing, including sandboxing and probing.

Provider and privacy options

The site lists support for Ollama, OpenAI, Anthropic Claude, OpenRouter, Google Gemini, DeepSeek, Groq, Together AI, LM Studio, llama.cpp, vLLM and compatible OpenAI chat API endpoints. The CLI can run fully offline with a local Ollama model; the site says this mode makes no network calls and sends no telemetry.

AgentSeal says prompts travel directly from the CLI to the user's LLM provider. Prompts and model configurations synced to the dashboard are encrypted at rest with Fernet (AES-256).

Reports and automation

For CI workflows, AgentSeal supports SARIF 2.1.0, JUnit XML, GitHub Actions summaries and policy-as-code rules for gating pull requests. Report formats also include JSON and PDF. The available attack categories include prompt extraction, instruction injection, data exfiltration and MCP tool poisoning; target systems include system prompts, AI agents, HTTP endpoints, MCP servers, RAG pipelines and multimodal systems.

Pricing

AgentSeal offers a free plan and a Pro plan. The free plan costs 0.00 USD per free and includes 30 basic probes, the MCP registry, Guard and Watch. The Pro plan's price is not listed; its stated features include 311 probes, runtime MCP scanning with OAuth, PDF export and the dashboard.

Platforms

AgentSeal is listed for API, Linux, macOS, web and Windows. The CLI is installable with pip, and an npm wrapper is available for Node projects and CI pipelines. The CLI requires Python 3.10 or higher; the npm wrapper runs the Python CLI.

Who it's for

AgentSeal is aimed at people responsible for the security of AI agents, prompts and MCP servers, particularly those who need both adversarial checks and ongoing monitoring. Its CLI, local Ollama option, CI output formats and pull-request gating rules may suit teams incorporating checks into development workflows. The dashboard adds centralized scan results, prompt management, monitoring and GitHub integration settings.

The free plan offers a way to start with basic probes and local monitoring tools. Teams needing the full 311-probe coverage, OAuth-enabled runtime MCP scanning, PDF exports or dashboard access should note that these features are listed under Pro, for which no price is provided.

Pros and cons

Pros

• Combines prompt probing, local configuration checks, MCP server testing and continuous monitoring.
• Supports offline CLI use with a local Ollama model, according to the site.
• Provides SARIF, JUnit XML, JSON, PDF and GitHub Actions summary reporting, with policy rules for pull-request gating.
• Offers a free plan with 30 basic probes, the MCP registry, Guard and Watch.

Cons

• The listed free plan includes 30 basic probes, while the 311-probe coverage is associated with Pro.
• The Pro price is not listed, making it difficult to assess the cost of its additional features from the available information.
• The npm wrapper depends on the Python CLI, which requires Python 3.10 or higher.

Alternatives

Other tools to consider include OpenSecureAI Scanner, Promptfoo, RedAmon, F5 BIG-IP APM, NVADER, ProofLayer, Confident AI and Darkhunt AI Security.

Verdict

AgentSeal brings together prompt testing, machine-local agent checks, MCP server scanning and continuous monitoring, with options for offline CLI use and CI integration. Its free plan includes a useful set of basic tools, while several advanced capabilities—including the broader probe coverage and runtime MCP scanning with OAuth—are assigned to Pro. Because the Pro price is not listed, prospective users should weigh those plan boundaries against their requirements before choosing it.