VulX Watch MCP

Independent security review for AI-built apps, as a watch, not a patch. VulX connects, watches a GitHub repository, and answers what the last reading found. It deliberately returns no patch, no diff, and no code to apply - the review is the product, and remediation stays with the team.

Server type: Hosted, remote (Streamable HTTP)
Endpoint: https://mcp.vulx.ai/mcp
Auth: OAuth (GitHub), no API key
Registry: ai.vulx/watch (v1.0.0, active)
Tools: 3 (watch_repo, security_status, + review controls)
Built by: VulX (vulx.ai)

Why This Matters for Operators

Teams shipping AI-built applications need a security signal that is independent of the code generator. VulX gives agents a read-only watch on a repo.

First, independent reading. The reviewer is separate from the code that was written, which matters exactly when the code was generated by an AI that also reviewed its own output.

Second, watch semantics. watch_repo puts a repository under watch and starts a reading; security_status reads what the last completed review found, read-only, without surprise costs. The agent can integrate this into a CI or triage loop.

Third, no remediation drift. Because VulX never returns patches, the team stays in control of fixes, and the agent's job stays advisory.

Tools and Capabilities

Tool	What it does
watch_repo	Put a GitHub repo under watch and start a reading (owner/name or github.com URL)
security_status	Read what the last completed review found. Read-only
Review controls	Manage watch state and review lifecycle

Verification (Aug 31, 2026)

• Endpoint responded to a JSON-RPC initialize attempt with {"detail":"Missing bearer token"} - the gateway is live and OAuth-gated (GitHub), confirming the auth posture before token exchange.
• Submission (#3863) documents the three tools, the no-patch policy, and the active registry record ai.vulx/watch v1.0.0.