What it does
BackBond Agent Scan checks the tool metadata of an MCP server before it is attached to an agent. It reads an exported tools/list manifest and returns one of three results: block, review, or no blocking finding. It reports findings and missing evidence, and it does not execute the proposed tools or probe a live agent. A no-blocking-finding result covers only the supplied metadata and does not establish runtime safety.
Operators would reach for it when they need to review a proposed MCP tool list before the agent can use it. A browser preview runs three synthetic examples and can check a manifest pasted locally, with no account and no upload. For larger inputs, other supported formats or offline work, a pinned command-line version is available, including a single-file build that runs without npm. Installing the bundled Agent Skill adds guidance for agents, but installing it does not run a scan.
Running the command-line tool requires a manifest saved as tools-list.json in the current directory; the scanner reads that file but does not create it. The publisher describes a checksum-verified path for hardened or network-isolated hosts, where the file is verified on a trusted machine and then transferred. A failed download is reported as a scan that did not run, not as a clean result.
Limitations and requirements
Only static metadata is inspected; runtime behaviour is not assessed. The browser preview is limited to 200 tools and 256 KiB of supplied metadata. The scanner does not enforce attachment or provide insurance coverage. A non-blocking result is not a safety determination. The command-line workflow needs an exported tools/list file and network access for npm installation unless the standalone file is used. The separate stdio gate is not included in the scanner package and does not verify runtime behaviour. The skill installer is separate software that fetches the public repository and has its own network and telemetry behaviour.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
BackBond Agent Scan
Vet MCP tool manifests before attachment; returns block, review, or no blocking finding.
v0.6.3As shown on registry.modelcontextprotocol.io when we last read it