# AgentSecurityLens MCP Security Trust Check [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/professor2k8/agent-security-lens  
**GitHub Stars:** 0  
**npm Downloads (last month):** 522  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/agentsecuritylens-mcp-security-trust-check

## Description
MCP security trust-check for agents before installing MCPs, Skills or tools.

## Tools
Capabilities this server exposes over MCP:

- **get_install_policy** — returns the current Agent execution policy and records a minimal first-call activation event.
- **get_intelligence_status** — reports online or local intelligence status for MCP clients.
- **review_before_install** — evaluates a proposed component and installation context.
- **check_component** — retrieves known component intelligence.
- **recommend_alternatives** — returns evidence-backed alternatives and mitigations.
- **submit_unknown_component** — submits public metadata for research.
- **get_research_status** — checks whether an unknown component has been reviewed.
- **report_install_outcome** — reports the result of an Agent action.
- **submit_decision_feedback** — reports useful, incorrect, or incomplete decisions.
- **discover_workspace** — discovers Agent, MCP, and Skill environments.
- **scan_workspace** — performs a local fallback assessment.

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

```json
"mcpServers": {
  "agentsecuritylens-mcp-security-trust-check": {
    "command": "npx",
    "args": ["-y","agent-security-lens"]
  }
}
```

## Documentation

## What the AgentSecurityLens MCP Security Trust Check MCP server does

AgentSecurityLens MCP Security Trust Check MCP server provides a pre-installation review layer for autonomous agents. It evaluates MCP servers, Skills, tools, prompt bundles, workflows, and memory components using the proposed source, installation command, intended use, and requested permissions. The result is not a malware verdict. Instead, it describes observable risk signals, controls to apply, available alternatives, and whether automatic installation is permitted.

The service distinguishes between decisions such as `allow`, `allow_with_restrictions`, `ask_user`, and `avoid`. Unknown or incompletely reviewed components do not receive automatic-install authorization. Agents are expected to follow the returned decision contract and structured actions rather than infer policy from descriptive text.

## How it works

After connecting, an agent should call `get_install_policy`, then use `review_before_install` for each planned component. A review can include a component name and type, source URL, install command, planned use, requested permissions, and whether unknown metadata may be submitted. The response can include a trust score, risk signals, a safe-install plan, required user confirmation, and a one-step action.

Online mode uses public component metadata and an online intelligence API. If that service is unavailable, the package can use a local fallback containing basic rules and a limited public intelligence baseline. The online catalog includes broader coverage, while automatic assessments still cannot authorize automatic installation. `get_intelligence_status` exposes the current online or local state.

AgentSecurityLens MCP Security Trust Check MCP server also supports a review loop after installation. Agents can report the result with `report_install_outcome`, submit useful or incorrect decision feedback, and check whether an unknown component has later been researched. `discover_workspace` and `scan_workspace` help identify and assess local Agent, MCP, and Skill environments.

## Setup and configuration

Node.js 20 or newer is required. A standard stdio configuration launches the package with `npx -y agent-security-lens`. The repository provides installation guidance for Claude Desktop, Cursor, Codex-style setups, and generic MCP configurations.

Online intelligence is enabled by default. `ASL_API_URL` can select an endpoint, while `ASL_API_URLS` accepts comma-separated endpoints that are tried in order. To use the local fallback explicitly, set `ASL_MODE=local`. The first policy call may record a minimal activation event when online mode is enabled; `ASL_DISABLE_USAGE_TELEMETRY=1` disables that usage event.

Do not submit private source code, file contents, secrets, passwords, tokens, cookies, private prompts, memories, or customer records. Online lookup is intended for public component metadata.

## Tools and capabilities

- Check the active installation policy and intelligence availability.
- Review a proposed component before installation.
- Retrieve known component intelligence and suggest alternatives or mitigations.
- Submit public metadata for unknown components and check research status.
- Report installation outcomes and decision feedback.
- Discover relevant local environments and run a local fallback assessment.

AgentSecurityLens MCP Security Trust Check MCP server includes a public reviewed baseline, curated fallback information, and online access to an expanded catalog. Its scoring method is documented in the repository's Agent Component Safety Standard.

## Limitations and notes

The server evaluates evidence and installation context; it does not declare a component malicious without supporting evidence. Local mode has less coverage than online intelligence. Automatic assessments cannot authorize automatic installation, and unknown records should be treated conservatively.

The package does not include the complete private intelligence database. Agents should apply every item in `safe_install_plan`, stop for confirmation when required, and avoid automatic installation when `automatic_install_allowed` is false. The project is licensed under Apache-2.0.

_Full upstream README: https://allmcps.com/mcp/agentsecuritylens-mcp-security-trust-check/readme_