Helixar Security

What Helixar Security MCP server does

The Helixar Security MCP server provides three security-focused MCP tools for different inputs:

• helixar_inspect_mcp examines an MCP server URL or raw manifest JSON using Sentinel detection rules. Results include a risk score, findings, and a security brief. Quick mode is public and checks the top eight rules; deep mode runs all 26 rules when an API key is supplied in the tool arguments.
• helixar_hdp_validate checks an HDP delegation chain against IETF draft draft-helixar-hdp-agentic-delegation-00. It reports issues such as scope escalation, excessive chain depth, expired hops, and absent signatures. Results cite the draft and its Zenodo DOI.
• helixar_releaseguard invokes the Helixar-AI/ReleaseGuard command-line binary against dist/ or release artifacts. Quick checks look for secrets, metadata exposure, and license gaps. Its deeper operations can run the harden pipeline, including fixing, obfuscating, signing, and attesting artifacts.

How it works

The project is a TypeScript ESM application built for Node 20 or newer and uses the official MCP SDK. The local entry point communicates over stdio. A separate Cloudflare Workers adapter exposes a stateless Streamable HTTP endpoint at https://mcp.helixar.ai/mcp.

The remote deployment currently exposes helixar_inspect_mcp and helixar_hdp_validate. helixar_releaseguard remains local because it starts an external Go binary through child_process. URL inspection includes SSRF protections, with separate guards for Node and Cloudflare Workers. Security narration uses the Anthropic SDK when configured and has a deterministic fallback when no API key is available.

Setup and configuration

For local use, clone the repository, install dependencies, build the project, and run the stdio entry point. The resulting dist/server.js can be configured as the command for Claude Desktop or Claude Code. A local process provides all three tools.

The hosted endpoint is public and does not require authentication in v1. Deep inspection and advanced ReleaseGuard modes signal access through a non-empty api_key field in tool input arguments. The README describes OAuth 2.0 and dynamic client registration as a later Phase 8 item; no environment variable is specified as required for startup.

Tools and capabilities

The Helixar Security MCP server supports manifest analysis from either inline JSON or a URL. Quick inspection covers eight rules, while deep inspection covers 26. HDP validation uses nine validation rules and identifies delegation-chain problems rather than scanning general MCP behavior. ReleaseGuard can check artifacts locally and, in deep mode, run its broader hardening workflow, provided the releaseguard binary is available on PATH.

Limitations and notes

Remote use does not provide the ReleaseGuard tool. The hosted service is open without authentication in the stated v1 configuration, and deep-mode API-key validation is described as a pre-OAuth mechanism rather than full OAuth. Internal detection methodology, thresholds, and implementation details are not exposed; outputs provide rule IDs, severity categories, public-safe detection categories, and remediation guidance. The earlier helixar_triage_alert tool is not part of the current release.