agent-bom
Open security scanner and self-hosted control plane for AI, MCP, and cloud. One evidence model — run scans in your environment, centralize findings, govern in your VPC.
An open scanner and control plane for AI and cloud risk
agent-bom scans repositories, containers, clusters, and MCP servers, then correlates the evidence into one finding and graph model for prioritized investigation.
Scanning and correlation
agent-bom scans repositories, developer endpoints, images, clusters, cloud and data platforms, MCP servers, and runtime activity, then normalizes the evidence into one Finding plus UnifiedGraph model. A blast radius view connects a package finding to the AI surfaces that can reach it: a package vulnerability leads to an MCP server, which leads to a connected agent, credential environment names, and reachable tools. The terminal report shows this attack path for human review, and MCP clients can query the same evidence through an exposure_paths field and use a should_i_deploy call for a bounded pre deployment verdict.
Deployment and surfaces
The same deterministic scanner runs on a workstation, in CI, in Docker or Kubernetes, or inside a self hosted control plane, producing findings, SARIF, SBOMs, HTML reports, and graph exports. A self hosted loopback control plane is started with agent-bom serve after installing the UI extra. In MCP server mode the tool exposes 81 MCP tools, 6 resources, and 8 workflow prompts, all read first so discovery and analysis never mutate a scanned target. The project supports Python 3.11 through 3.14 and documents Docker, Helm, EKS, and Snowflake deployment paths.
Trust and evidence
The project follows a read only discovery default, with runtime write decisions kept separate and explicit. Credentials are written only where stored, encrypted at rest, and never returned by API responses. API and control plane routes are tenant scoped and auth protected outside local mode. Missing evidence is shown as unavailable or partial rather than converted into a factual zero, and public examples use deterministic synthetic identifiers only. The README states the tool is Apache 2.0 licensed.
Editorial conclusion
agent-bom is a Python project released under the Apache 2.0 license and ships as a PyPI package, a container image, and a Helm chart, with a GitHub Action for gating continuous integration.
Community notes