AgentScan

AgentScan is a cloud service for automated security testing of LLMs, RAG pipelines, and autonomous agents through an API. Its scans cover attack patterns such as prompt injection, data exfiltration, jailbreaks, and MCP attacks. Results include a 0–100 risk score, severity breakdown, remediation advice, and evidence export in JSON or PDF. The web dashboard can detect agent formats, show scan progress and errors, adjust throttling, and download PDF reports. The REST API queues scans asynchronously, returns a scan ID, and supports completion webhooks. AgentScan maps vulnerabilities to OWASP LLM Top 10 risks and describes connections to EU AI Act articles and NIST AI RMF functions. Its compliance documentation says its OWASP mapping does not test Model Denial of Service or Model Theft. The Free plan is free and includes five monthly scans across 185 attack vectors and five categories. Starter costs $29 per month and includes 30 scans; paid plans start at $29/mo. The terms require customers to scan only agents they own or have explicit written authorization to test.

AgentScan combines API-based scanning with dashboard reporting and framework mappings. Its documented OWASP coverage has exclusions, and the terms require authorization for every agent tested.

It may suit teams that need API-based security scans for LLMs, RAG pipelines, or autonomous agents. Use is limited to systems the customer owns or is explicitly authorized to test.

Overview

AgentScan is a cloud service for automated security testing of LLMs, RAG pipelines and autonomous agents, with API and web access. It is best suited to teams that need repeatable scans and actionable reports; its tiered scan limits and required authorization make it a less natural fit for unrestricted experimentation.

Key features

AgentScan covers prompt injection, jailbreak, data exfiltration and other attack classes. Its full catalog reaches 412 attack vectors across 19 categories, including MCP attacks, but that breadth is reserved for Enterprise: Free, Starter and Pro cover 185 vectors in five categories, 300 in 12 and 348 in 14, respectively. The OWASP LLM Top 10 mapping also excludes Model Denial of Service and Model Theft, so the mapping should not be treated as comprehensive coverage of those risks.

Reports pair a 0–100 risk score with severity, remediation guidance and evidence, and can be exported as JSON or PDF. Free includes JSON results; PDF reports start with Starter. The dashboard detects agent format, tracks scan progress and errors, allows live throttle adjustments and supports PDF downloads. These features make AgentScan useful for teams that need findings they can review and share, though the lower tiers limit how much of its attack catalog those findings represent.

The documentation covers 12 agent types, from generic HTTP and OpenAI-compatible endpoints to tools such as Dify, Flowise, n8n, OpenClaw, AutoGPT, OpenHands, LangChain and Manus. Integration examples include OpenAI, Groq and LiteLLM. Its REST API queues scans asynchronously, returns a scan ID and supports completion webhooks, a practical fit for automated workflows. Framework mappings connect findings to OWASP LLM Top 10 risks and describe links to EU AI Act articles and NIST AI RMF functions.

Security controls include Argon2id-hashed passwords, HMAC-SHA256-hashed API keys, TLS connections and SSRF protection for outbound requests. Results remain in the account until deleted; account data is permanently removed within 30 days of account deletion. User data is stored on Hetzner servers in Germany. The terms require customers to scan only agents they own or have explicit written authorization to test, a firm boundary for consultants and security teams working across client environments.

Pricing

AgentScan is freemium, with paid plans from $29/mo. Free costs nothing (billed $0/month) and allows five scans per month, 185 attack vectors and five categories, with agent-format auto-detection and JSON results. It is useful for occasional evaluation, but the scan cap and narrower coverage constrain recurring security work.

Starter costs $29 a month (billed $29/month). It raises the allowance to 30 scans per month, 300 vectors and 12 categories, and adds PDF reports, a full test audit trail and three concurrent scans. This is the clearest fit for a small team that wants regular scans and shareable reports, though it still omits a substantial portion of the full catalog.

Pro costs $99 a month (billed $99/month) and includes 100 scans per month, 348 vectors across 14 categories, custom secrets monitoring, configurable throttling, PDF reports, webhooks and priority support. It suits teams integrating scans into a more regular workflow, but remains capped below Enterprise coverage. Enterprise has custom pricing and includes unlimited scans, all 412 vectors and 19 categories, MCP, State Injection and PII coverage, plus an SLA and priority support. Choose it when the stated scan limits or full catalog are necessary.

Platforms

AgentScan is cloud-hosted and supports API and web access. It does not offer a self-hosted deployment mode, which may rule it out for teams that require locally hosted testing.

Who it's for

AgentScan is a strong match for engineering and security teams responsible for authorized LLM, RAG or agent deployments, especially when they want evidence-backed results, workflow automation and framework-oriented reporting. Its limited monthly quotas at lower tiers and cloud-only deployment make it less suitable for high-volume scanning on a tight budget or environments that require self-hosting.

Pros and cons

• Pro: Reports combine risk scoring, severity, remediation advice and evidence, giving teams useful material for prioritizing fixes.
• Pro: The API supports asynchronous scans and completion webhooks, while the dashboard tracks progress and errors.
• Pro: Agent and integration examples span common frameworks and providers, including Dify, n8n, OpenAI and Groq.
• Con: Free is limited to five scans monthly and 185 vectors; even Pro does not include the full 412-vector catalog.
• Con: The OWASP mapping does not test Model Denial of Service or Model Theft.
• Con: Cloud hosting and the authorization requirement may not fit teams needing self-hosted testing or broader testing rights.

Alternatives

For other options, browse AI Security Testing Tools. Consider F5 BIG-IP APM if its broad platform availability and free trial better suit your evaluation needs. Project Moonshot is a free alternative. Crucible offers a free plan and Linux or self-hosted availability. OpenSecureAI Scanner may suit teams that want a larger stated API request allowance: its Pro plan is $49 a month (billed $49/month) for 50,000 API requests monthly. PromptGuard offers a free plan with 20,000 scans monthly, one API key and one project, and a Team plan at $19 a month. RedForge is a free, Apache 2.0 open-source option installable with pip. AIRTA Red Team is free and open source, with a bring-your-own-LLM-API-key model. AIRTaaS offers a free web plan.

Verdict

Choose AgentScan if your team needs authorized, repeatable cloud scans with evidence-rich reporting and API automation. Its strongest case is the combination of agent coverage, actionable reports and framework mappings; look elsewhere if you need self-hosting, unrestricted scan volume at a lower tier, or OWASP mapping coverage for Model Denial of Service and Model Theft.