AgentSeal
agentseal.org
- Free plan
- Open source
- Windows
- macOS
- Linux
- Web
- API
- Self-hosted
What AgentSeal is
AgentSeal is an open-source security scanner and toolkit for testing AI agent prompts, auditing MCP servers, and detecting agent vulnerabilities. Its scan command uses adversarial probes to look for prompt extraction and injection weaknesses. Guard scans agent configurations and skill files on a machine, while Watch monitors those files and MCP configurations for changes. Scan-MCP runs MCP servers in a sandbox and tests tool behavior with adversarial payloads; OAuth is supported for authenticated remote servers. The site lists providers including Ollama, OpenAI, Anthropic Claude, Google Gemini, and compatible OpenAI chat API endpoints. The CLI can run offline with a local Ollama model, with zero network calls and zero telemetry according to the site. A dashboard provides scan results, prompt management, security monitoring, and GitHub integration settings. CI output includes SARIF 2.1.0, JUnit XML, GitHub Actions summaries, and policy-as-code rules for pull-request gating. The free plan costs nothing and includes 30 basic probes, an MCP registry, Guard, and Watch. CLI installation requires Python 3.10 or higher; an npm wrapper is also available.
Who it suits
It suits teams and developers checking AI prompts, agent configurations, and MCP servers. CI output and policy rules may be useful for teams that gate pull requests on scan results.
Good to know, and what to watch
- Guard and Watch cover agent files and MCP configurations.
- Scan-MCP tests server behavior in a sandbox.
- Can run offline with a local Ollama model.
- CI outputs include SARIF, JUnit XML, and GitHub summaries.
- Free plan includes 30 basic probes.
- CLI requires Python 3.10 or higher.
- The free plan includes 30 basic probes.
- The npm wrapper shells out to the Python CLI.
Plans
Every plan and priceFree
Pro
Free plan
Getting started
- Install the CLI with pip
- Use Python 3.10 or higher
- Install the npm wrapper for Node projects or CI pipelines if needed
- Select a listed LLM provider, or use a local Ollama model offline
- Run prompt, configuration, or MCP scans
- Connect the dashboard for results, monitoring, and GitHub settings
Limits to check
The Free plan includes 30 basic probes, the MCP registry, Guard, and Watch. The Pro plan expands coverage to 311 probes and adds runtime MCP scanning with OAuth, PDF export, and the dashboard.
The Club's review
Overview
AgentSeal is a security toolkit for finding weaknesses in AI agent prompts, configurations, and MCP servers. It is best suited to developers and security teams building agent workflows who want repeatable checks, CI gates, or local-only scanning. Its mix of prompt probes and ongoing machine-level monitoring is compelling, though the free plan’s 30 basic probes are a limited starting point for broader coverage.
Readers comparing this category can browse AI Red Teaming Tools.
Key features
Prompt testing and monitoring
The scan command sends adversarial probes at system prompts to uncover extraction and instruction-injection weaknesses. The FAQ describes 311 probes spanning extraction, injection, MCP tool poisoning, RAG poisoning, and multimodal attacks. That breadth makes AgentSeal relevant beyond prompt-only reviews, but the free plan includes just 30 basic probes; Pro is the tier associated with 311.
Guard scans agent configurations and skill files on a machine, while Watch monitors those files and MCP configurations for changes. Together they extend the product beyond point-in-time checks, which is useful for teams that need continuing oversight rather than occasional manual scans.
MCP, providers, and privacy
Scan-MCP runs MCP servers in a sandbox and tests tool behavior with adversarial payloads. OAuth support allows testing authenticated remote servers, a meaningful addition for teams whose MCP services are not public. The MCP registry also applies a seven-stage security pipeline, including sandboxing and probing, before listing a server.
The CLI supports Ollama, OpenAI, Anthropic Claude, OpenRouter, Google Gemini, DeepSeek, Groq, Together AI, LM Studio, llama.cpp, vLLM, and compatible OpenAI chat API endpoints. Prompts go directly from the CLI to the user’s LLM provider; dashboard-synced prompts and model configurations are encrypted at rest with Fernet (AES-256). The CLI can also run fully offline with a local Ollama model, with zero network calls and zero telemetry, which is a strong fit for sensitive workflows that cannot send prompts to hosted services.
Automation and workflow
AgentSeal supports SARIF 2.1.0, JUnit XML, GitHub Actions summaries, and policy-as-code rules for gating pull requests. Its continuous monitoring and CI outputs make it useful for teams that want security findings to enter existing build workflows instead of relying on a separate review process. The dashboard adds scan results, prompt management, security monitoring, and GitHub integration settings.
The CLI installs with pip; an npm wrapper for Node projects and CI pipelines shells out to the Python CLI, so it does not remove the Python dependency. The CLI requires Python 3.10 or higher. Support typically responds within 24 hours, with an email address for urgent matters.
Pricing
AgentSeal is freemium, with a Free plan at no cost. It includes 30 basic probes, the MCP registry, Guard, and Watch. That is a useful way to try core scanning and monitoring, but the probe cap makes it less suitable as the sole option for teams needing the full described scan coverage.
The Pro plan has custom pricing and includes 311 probes, runtime MCP scanning with OAuth, PDF export, and the dashboard.
Platforms
AgentSeal is available on API, Linux, macOS, web, and Windows. The CLI’s Python 3.10-or-higher requirement matters most to developers integrating local scans or CI jobs; the npm wrapper still invokes that Python CLI.
Who it's for
AgentSeal is a good fit for AI developers and security teams who need to probe prompts, inspect MCP behavior, and monitor local agent files or configurations over time. Its offline Ollama mode and pull-request gating also suit teams with strict privacy requirements or established CI processes. It is less suited to buyers who need a published Pro price before evaluating cost, or who expect the free tier to provide all 311 probes.
Pros and cons
Pros
- Combines adversarial prompt testing, machine-level Guard and Watch checks, and sandboxed MCP scanning in one toolkit.
- Offline Ollama operation avoids network calls and telemetry, while dashboard-synced data is encrypted at rest.
- SARIF, JUnit, GitHub Actions summaries, and policy rules support security checks in CI and pull-request workflows.
- OAuth support for remote MCP servers and a seven-stage registry review add coverage beyond local prompt scans.
Cons
- The free plan is capped at 30 basic probes, well below the 311 probes described for Pro.
- Pro pricing is custom, so teams cannot compare a fixed subscription cost from the plan price alone.
- The CLI requires Python 3.10 or higher, including when using the npm wrapper.
Alternatives
Choose OpenSecureAI Scanner if a published paid tier is more important: its Pro plan is $49 a month, with 50,000 API requests per month, 120 requests per minute per key, hosted AI remediation and LLM Gateway completions, and priority email. Choose Promptfoo if you want a free community tier with 10k red-team probes per month, broad model-provider integrations, and local or self-hosted operation.
RedAmon is an alternative for readers seeking a free, MIT-licensed self-hosted Docker stack for personal or commercial use. Giskard may suit teams looking for an open-source library with local deployment and basic LLM vulnerability scanning and RAG evaluation. NVADER is another freemium option, available on the web. Rogue offers a free tier for personal and internal use, with commercial hosting priced separately. Darkhunt AI Security offers free credits and full platform access on its free plan. F5 BIG-IP APM is a paid alternative with a free trial and perpetual licensing across virtual editions, hardware, or hybrid environments.
Verdict
Choose AgentSeal if you need to connect adversarial prompt and MCP checks with ongoing machine monitoring, CI gates, or offline local scanning. Its strongest case is the combination of those workflows with broad provider support and privacy-conscious operation. Look elsewhere if a clear Pro price is essential or if 30 free probes cannot cover your evaluation needs.
The verdict
AgentSeal combines prompt scanning, machine-level checks, and MCP testing, with options for offline use and CI workflows. The free plan is a starting point, while CLI users need Python 3.10 or higher.
Questions people ask
Is AgentSeal open source?
Yes. AgentSeal is described as an open-source security scanner and toolkit.
What does the free plan include?
The Free plan costs nothing and includes 30 basic probes, the MCP registry, Guard, and Watch.
Which platforms are supported?
AgentSeal lists API, Linux, macOS, web, and Windows support. The CLI requires Python 3.10 or higher.
Can it run without network calls?
Yes. The CLI can run fully offline with a local Ollama model, with zero network calls and zero telemetry according to the site.
What reports can it export?
Listed outputs include JSON, SARIF 2.1.0, JUnit XML, PDF, and GitHub Actions step summaries.
What the maker publishes
- API and SDKs
- AgentSeal documents a Python AgentValidator API, a TypeScript AgentValidator API, and a REST dashboard API. source
- Attack coverage
- Pro adds 45 MCP-poisoning, 28 RAG-poisoning, and 13 multimodal probes plus behavioral genome mapping. source
- CI/CD outputs
- The CLI can emit terminal, JSON, SARIF, and JUnit results and enforce score or severity gates. source
- CI integrations
- AgentSeal supports SARIF 2.1.0, JUnit XML, GitHub Actions summaries, and policy-as-code rules for gating pull requests. source
- Dashboard
- The dashboard provides scan results, prompt management, security monitoring, and GitHub integration settings. source
- Dashboard
- The web dashboard provides scan history, trust-score trends, encrypted prompt storage, agent monitoring, API-key management, and GitHub integration. source
- Deterministic results
- The scanner uses hardcoded probes, canaries, and n-gram matching rather than an LLM judge, so repeated runs are reproducible. source
- Independent license concern
- Hysen Labs notes that FSL-1.1 is not OSI-approved and says the repository badge and metadata disagree, so users should read the license file before commercial use. source
- Independent project summary
- SourcePulse describes AgentSeal as a toolkit for dangerous-skill detection, poisoned-config scanning, prompt red-teaming, and live-system auditing. source
- Installation
- The CLI is installable with pip, and the site also provides an npm wrapper for Node projects and CI pipelines. source
- Installation
- The project publishes installation through pip and npm packages. source
- License
- The GitHub repository and PyPI metadata identify the license as FSL-1.1-Apache-2.0. source
- License discrepancy
- The official documentation describes the core scanner as MIT licensed, while the repository identifies FSL-1.1-Apache-2.0. source
- Limit: adoption
- RightAIChoice says community adoption is limited, making real-world effectiveness difficult to judge. source
- Limit: dashboard bug
- The public GitHub issue tracker shows an open bug titled inability to connect to the dashboard. source
- Limit: deployment
- RightAIChoice says there is no plug-and-play cloud version and users may need to run the CLI or integrate it themselves. source
- Limit: model costs
- RightAIChoice notes that prompt scanning can require paid API access to an external language-model provider. source
- LLM providers
- The site lists Ollama, OpenAI, Anthropic Claude, OpenRouter, Google Gemini, DeepSeek, Groq, Together AI, LM Studio, llama.cpp, vLLM, and compatible OpenAI chat API endpoints. source
- Machine guard
- Guard scans local agent skill files, MCP configurations, hidden payloads, and toxic data flows without an API key. source
- Machine protection
- Guard scans agent configurations and skill files on a machine, while Watch monitors those files and MCP configs for changes. source
- Maker identity
- LinkedIn identifies the organization name as AgentSeal and categorizes it under Technology, Information and Internet. source
- Maker size
- LinkedIn lists AgentSeal as self-employed, founded in 2026, with 2–10 employees. source
- MCP scanning
- Scan-MCP runs MCP servers in a sandbox and tests tool behavior with adversarial payloads; OAuth is supported for authenticated remote servers. source
- MCP scanning
- Scan-mcp connects to running MCP servers and analyzes tool definitions for poisoning, toxic flows, and rug pulls. source
Checked against the maker's pages on 15 Oct 2026.