Inspect an MCP tool manifest before attaching it to an agent

Описание и применение

Проверка метаданных MCP‑манIFEST перед подключением к агенту с помощью локального статического сканера позволяет обнаружить манипуляции выбором инструментов и блокировать потенциально опасные наборы.

Задача

Operators need to verify MCP tool manifests for malicious or misleading metadata before enabling them in an agent, to prevent unintended tool invocation or manipulation.

Результат

The workflow demonstrates that a local static scan can detect selection‑manipulation language in a tool manifest, blocking or flagging it before attachment. The example shows a BLOCK result with a high finding (BB013) for the synthetic manifest.

Рабочий процесс

• Download the tools-list.json and its .sha256 checksum using curl or PowerShell.
• Verify the checksum with sha256sum or Get-FileHash to ensure file integrity.
• Run the pinned Agent Scan CLI: npx -y @backbond/[email protected] vet-tools --stdin < tools-list.json (or equivalent PowerShell cmd).
• Interpret the exit code: 0 = no_blocking_finding, 1 = block, 3 = review.
• If blocked or review, investigate the findings and decide whether to modify or reject the manifest before attachment.
• If no_blocking_finding, proceed to attach the tools to the agent.
• Document the scan result and any operator decisions for audit purposes.

Доказательства

For this exact synthetic fixture, the verified result is BLOCK with one high finding, BB013, and exit code 1.

Источник

The pre-attachment gate returns block (exit 1), review (exit 3), or no_blocking_finding (exit 0).

Источник