MCP Server Pack

Security checks across static analysis, malware telemetry, and agentic risk

Overview

This skill is purpose-aligned as an MCP server pack, but it can enable powerful file, credential, database, cloud, memory, Docker, and npm integrations with limited scoping and provenance detail.

Review this skill before installing. Only enable the MCP servers you need, prefer self-hosting for sensitive data, pin and verify any Docker images or npm packages, restrict file mounts and credentials, and do not accept a default-all generated config without manual review.

Static analysis

No static analysis findings were reported for this release.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Risk analysis

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

If the generated config is accepted blindly, an agent could gain broad local and third-party service capabilities beyond what the user intended for a single task.

Why it was flagged

A read-only config generator can produce an MCP configuration for all servers by default, including file-write, account, and database access, without clearly defining approval gates, allowed paths, or containment.

Skill content
servers: ... description: "Server names to include (default: all)" ... "filesystem-secure" | File system access with sandbox (chroot) | Read/write within allowed roots ... "github" ... Requires GitHub token ... "postgres" ... Requires DB connection string
Recommendation

Generate configs only for specific needed servers, review every MCP command and permission, set explicit filesystem allowlists, and avoid using the default-all configuration.

What this means

Running the suggested configuration could execute unreviewed third-party code on the user's machine with access to mounted files or supplied credentials.

Why it was flagged

The self-hosted workflow relies on remote Docker images or npm packages, but the artifacts do not pin versions or digests and provide no source or install spec for review.

Skill content
The skill downloads Docker images or binaries ... "command": "docker", "args": ["run", "-i", "--rm", "-v", "/path/to/allowed:/data", "openclaw/mcp-filesystem-secure"] ... "command": "npx", "args": ["-y", "mcp-github"]
Recommendation

Use pinned image digests and package versions, verify publishers and source code, run containers with least privilege, and do not auto-run generated Docker or npx commands without review.

What this means

A misconfigured token or connection string could expose private repositories or database contents to the MCP server.

Why it was flagged

GitHub tokens and database connection strings are expected for these integrations, and the GitHub guidance is read-scoped, but these credentials still grant access to private account or database data.

Skill content
"github" | GitHub API integration (issues, PRs, repos, search) | Requires GitHub token ... "postgres" | PostgreSQL read-only queries | Requires DB connection string ... Create a fine-grained PAT with `issues:read`, `pull_requests:read`, `repo:status` scopes.
Recommendation

Use least-privilege, read-only credentials; avoid broad GitHub PAT scopes; use a restricted database user; and rotate credentials if they are placed into generated configs.

What this means

Sensitive or incorrect information stored in memory could persist and influence later agent behavior.

Why it was flagged

The skill explicitly includes a persistent memory server, which is purpose-aligned but can retain information across tasks and be reused later.

Skill content
"memory-enhanced" | Memory server with WAL + compaction survival | Persistent JSON store
Recommendation

Store only information you are comfortable retaining, review or clear memory periodically, and avoid placing secrets in persistent memory.

What this means

Data sent through cloud-hosted MCP servers may leave the local environment and be processed by the provider's infrastructure.

Why it was flagged

The cloud option sends MCP traffic to a hosted service. This is disclosed and TLS is mentioned, but it creates an external data boundary users should understand.

Skill content
We host the MCP servers on our infrastructure ... You get a unique connection URL (wss://mcp.openclaw.ai/server) ... Is data sent to cloud? For cloud servers yes, but encrypted in transit (TLS).
Recommendation

Use self-hosting for sensitive data, confirm the provider and privacy terms, and avoid routing secrets or private files through cloud-hosted servers unless necessary.