Comparison
How mcpprobe compares
There are three kinds of tool in this space: scanners that read MCP servers and configs, gateways that enforce policy while agents run, and general web scanners. This page lists the ones we know of, described from their own documentation, so you can pick what fits. Often that is more than one.
Last checked against each project's README and docs, linked below. “Not published” means we found no public price. Corrections: hello@mcpprobe.tech.
MCP scanners
Look at an MCP server or a machine's MCP setup and report what they find. None of them sit between the client and the server.
| Tool | What it does | Passive or active | CI support | Hosted | Licence | Price |
|---|---|---|---|---|---|---|
| mcpprobe | Tests one MCP server you run (HTTP endpoint or local stdio command) the way a client would: authorization, transport, tool metadata, rug-pull baseline, errors and limits. Records the request and response behind each finding. | Passive by default; never calls a tool. Active checks are gated and not implemented yet | GitHub Action, SARIF, --fail-on | CLI runs locally; optional hosted plans | Apache-2.0 | CLI free; hosted Team £49/month, Business £199/month |
| Snyk Agent Scan (formerly mcp-scan) Detailed comparison | Finds the agents, MCP servers and skills configured on a machine, fetches tool descriptions, and scores risks such as prompt injection, untrusted content, private data and destructive capabilities. | Reads client configs and connects to the servers they list to fetch tool descriptions; interactive runs ask before contacting each server | --ci flag | Analysis runs on the Agent Scan API (needs a Snyk API token); background fleet mode reports to Snyk Evo | Apache-2.0 | Not published |
| Cisco AI Defense mcp-scanner Detailed comparison | Scans tools, prompts, resources and instructions with YARA rules, an LLM analyser and the optional Cisco AI Defense API; also source-code behavioural analysis, PyPI and npm package scans, pip-audit and VirusTotal lookups. | Connects to servers to read metadata, or analyses JSON, source and packages statically; package code is never executed | Offline mode over pre-generated JSON, described as for CI/CD | Runs locally or as a REST API you host; AI Defense analyser is a Cisco service | Apache-2.0 | Scanner free; AI Defense subscription price not published |
| Ramparts | Scans MCP servers (HTTP, SSE, stdio) and agent skill files with YARA rules and LLM analysis, checks OSV.dev for known CVEs, and fingerprints configs and tools against a baseline to flag rug pulls. | Connects to servers to read metadata; analyses skill files on disk | SARIF 2.1.0; documented for gating pull requests | Runs locally; optional server mode you host | Apache-2.0 | Free (open source) |
| Nova Proximity | Discovers an MCP server's tools, prompts and resources, analyses agent skills, and evaluates them with NOVA pattern rules plus LLM evaluation. | Connects to servers to read metadata; analyses skill files | Not documented | Runs locally | GPL-3.0 | Free (open source) |
| MCP-Shield | Scans the MCP servers in local client config files for tool poisoning, exfiltration channels and cross-origin escalation; optional Claude analysis with your own API key. | Reads client configs and connects to the servers they list | Not documented | Runs locally | MIT | Free (open source) |
MCP gateways and proxies
Sit in the request path between clients and servers and enforce policy at runtime. They are not scanners, and they complement one.
| Tool | What it does | Passive or active | CI support | Hosted | Licence | Price |
|---|---|---|---|---|---|---|
| Docker MCP Gateway | Runs MCP servers in isolated containers behind one gateway. Verifies signatures of images in Docker's mcp/ namespace, screens tool arguments and responses for secret-like values by default, can restrict network egress, and logs tool names per call. | Runtime enforcement in the request path | Not applicable (runtime) | Runs locally or wherever you deploy it | MIT | Free (open source) |
| agentgateway | Proxy for MCP and A2A traffic with JWT, API-key and OAuth auth, CEL-based RBAC, rate limiting, TLS, guardrail filters and OpenTelemetry. A Linux Foundation project. | Runtime enforcement in the request path | Not applicable (runtime) | Self-hosted | Apache-2.0 | Free (open source) |
| IBM ContextForge | Registry and proxy that federates MCP, A2A and REST/gRPC behind one endpoint, with auth, rate limiting, a plugin framework and OpenTelemetry tracing. | Runtime enforcement in the request path | Not applicable (runtime) | Self-hosted (PyPI, containers, Kubernetes) | Apache-2.0 | Free (open source) |
| Lasso MCP Gateway | Proxies the servers in an mcp.json, masks tokens and secrets (basic plugin) or PII (Presidio plugin); the Lasso plugin adds prompt-injection and custom-policy checks through Lasso's API. | Runtime enforcement in the request path | Not applicable (runtime) | Runs locally; Lasso plugin calls Lasso's service | MIT | Gateway free; Lasso API price not published |
General web and API scanners
Web and API scanners that work on any HTTP service. They do not know MCP's rules unless you teach them.
| Tool | What it does | Passive or active | CI support | Hosted | Licence | Price |
|---|---|---|---|---|---|---|
| ZAP | General web application scanner and intercepting proxy. MCP-specific checks are not described in the sources we checked. | Passive and active scanning | Docker packaged scans, GitHub Actions, Automation Framework | Runs locally | Apache-2.0 | Free (open source) |
| Nuclei | Template-driven vulnerability scanner: you write or reuse YAML templates that send requests and match responses. | Whatever the templates do; can be active | Documented for CI/CD pipelines | Runs locally; ProjectDiscovery offers a cloud platform | MIT | CLI free; cloud has a free tier and paid seats from $200 per seat per month |
How to read this
- A scanner and a gateway answer different questions. A scanner tells you what a server exposes before you rely on it; a gateway limits what any server can do while an agent is using it. Running both is normal.
- Most MCP scanners read tool metadata. Of the tools listed, only mcpprobe documents tests of the MCP authorization and transport rules against a running server. It also does less in other areas: it does not walk client configs, scan agent skills, analyse source code or block anything at runtime.
- Stars, versions and features change quickly. Check the linked source before you rely on a cell, and tell us when one is out of date.
When not to use mcpprobe · Browse the checks · Run the quickstart
Sources
All fetched on 3 November 2026.
- mcpprobe-dev/mcpprobe README
- snyk/agent-scan README
- cisco-ai-defense/mcp-scanner README
- highflame-ai/ramparts README
- fr0gger/proximity README
- riseandignite/mcp-shield README
- docker/mcp-gateway README
- docker/mcp-gateway security model (docs/security.md)
- agentgateway/agentgateway README
- IBM/mcp-context-forge README
- lasso-security/mcp-gateway README
- zaproxy/zaproxy README
- ZAP automation documentation
- projectdiscovery/nuclei README
- ProjectDiscovery pricing