# BackBond Agent Scan [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/BackBond/agent-scan  
**GitHub Stars:** 1  
**npm Downloads (last month):** 327  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/backbond-agent-scan

## Description
Local static screening for MCP and AI-agent tool metadata before attachment

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "backbond-agent-scan": {
    "command": "npx",
    "args": ["-y","@backbond/agent-scan@0.6.3"]
  }
}
```

## Documentation

## What BackBond Agent Scan MCP server does

BackBond Agent Scan MCP server evaluates MCP tool metadata and supported agent configuration that already exists on the local machine. Its checks are static and deterministic: the scanner examines tool identities, descriptions, input schemas, and relationships within the same manifest. It can report a blocking finding, request operator review, or indicate that no configured blocking rule fired for the supplied metadata.

The package is published as `@backbond/agent-scan@0.6.3`. A standalone `agent-scan.cjs` build is also available for checksum verification and use without npm installation. Neither distribution runs proposed tools, probes a live agent, uploads traces, contacts a hosted service, or provides runtime safety evidence.

## How it works

Export the proposed server’s actual `tools/list` response to a local file such as `tools-list.json`, then pass it to `vet-tools` through standard input:

```bash
node agent-scan.cjs vet-tools --stdin < tools-list.json
```

The decision and exit status are:

- `no_blocking_finding`, exit `0`: no configured blocking rule fired on the supplied metadata.
- `block`, exit `1`: automatic attachment should stop.
- `review`, exit `3`: incomplete, ambiguous, or medium-severity evidence needs operator review.

The analysis requires recognizable tool names, a description or title, and one analyzable object input schema per tool for a non-blocking result. Opaque schema branches, conflicting aliases, mixed manifest dialects, non-ASCII identities, confusable names, directive-like example text, and inputs beyond the bounded analysis budget can prevent that result. Full JSON includes a canonical profile digest and actionable review items.

BackBond Agent Scan MCP server does not enforce attachment, approvals, audit behavior, or calls. A non-blocking result is not a runtime attestation, safety determination, or insurance decision.

## Setup and configuration

For a network-enabled development machine, run the pinned npm package with `npx`:

```bash
npx -y @backbond/agent-scan@0.6.3 scan
npx -y @backbond/agent-scan@0.6.3 vet-tools --stdin < tools-list.json
```

For restricted or network-isolated environments, download the standalone file and its SHA-256 checksum on a trusted host, verify them, transfer the verified file through an approved path, and run it locally with Node.js. The standalone build supports `scan`, `vet-tools`, and the `mcp` stdio entrypoint without network access.

The repository also provides a version-pinned Agent Skill. Installing that skill is separate from running a scan, and the skill installer’s own network and telemetry behavior should be reviewed before use in restricted environments. Manual installation places `SKILL.md` under `.agents/skills/agent-scan/`.

## Tools and capabilities

- Inspect supplied MCP tool manifests locally.
- Scan local agent configuration with the pinned `scan` command.
- Classify results as block, review, or no blocking finding.
- Produce full JSON with review items and a canonical profile digest.
- Emit compact `backbond-vet-summary/v1` JSON using `--summary-only` for JSONL collection.
- Run the scanner through its stdio MCP entrypoint.

Summary-only output retains the decision, version information, finding counts, coverage and rule histograms, review items, and prompt-template multiplicity. It omits identities, descriptions, artifact names, evidence pointers, and template hashes; it cannot be combined with `--json`, `--sarif`, or `--suggest-policy`.

## Limitations and notes

The scanner accepts local input and does not collect manifests from a Registry or call scanned tools. It does not assess runtime enforcement, approvals, audit behavior, traces, or actual execution. A failed download in a network-isolated setup must be treated as `scan_not_run`, not as a zero-finding result.

The separate Agent Scan stdio gate can start an authorized server and gate attachment and calls, but it is not included in this npm package, does not sandbox startup, and does not verify runtime behavior. Browser previews are synthetic and limited to supplied metadata; they do not replace the pinned CLI for larger inputs or offline workflows.

_Full upstream README: https://allmcps.com/mcp/backbond-agent-scan/readme_