Acquire AgentScore
AgentScore is a working, public security scanner for MCP packages on npm. It is for sale as an asset deal. This page tells you exactly what you would be buying and is honest about what you would not be.
The reason it is for sale is plain. It was built and run by one person as proof that MCP supply-chain risk could be scanned, scored, and disclosed continuously. It does that. It was never turned into a revenue business, and the standalone-scanner market has consolidated around larger players. The assets below are real, the scanning pipeline has run every day without a gap since March 31, 2026, and the right home for this is a team already building in MCP or supply-chain security.
What is included
The scanner engine and ruleset
A continuous scanner that pulls npm tarballs, walks their source for risk patterns (command injection, unsafe eval, hardcoded secrets, prompt-injection markers), classifies MCP tool capabilities, and scores each package. It carries a public precision-correction history: a dated changelog of its own false positives and false negatives and how each was fixed. That discipline is rare and is itself part of the asset.
An 86-day scan time-series dataset
Over 32,000 scans across more than 1,560 monitored MCP packages and 9,196 distinct package-versions, captured daily with no gap since March 31, 2026. Score, findings, dependency snapshot, and capability analysis per version over time. Current state can be re-derived by anyone with a scanner; the longitudinal history cannot be regenerated after the fact.
A public advisory archive with real provenance
70 published security advisories (2 critical, 41 high) with timestamped first-observation data, an RSS feed, and auto-filed GitHub issues. The provenance is the value: Redis pinned every MCP dependency in RedisInsight after a scan, Grafana pinned a package in k6-studio, and Agions and HomenShum shipped fixes. These are closed-completed issues in third-party repositories, not claims.
An official MCP registry listing and four npm packages
A listing in the official Model Context Protocol registry (io.github.Thezenmonster/agentscore, marked latest), plus four published npm packages including the MCP server client that lets any agent query the scanner. CI auto-publishes on tag push via OIDC trusted publishing.
The domain, the site, and the indexed surface
agentscores.xyz with around 960 indexed URLs: package reports, repo dossiers, the advisory feed, research, and case studies. A full Next.js application, the monitoring crons, the policy-gate API and GitHub Action, and the supporting infrastructure.
A live external consumer
As of June 2026 a real external agent has been using the scanner API in production, unprompted, querying package safety as part of its routine. One organic machine consumer, found with zero marketing. Evidence the core idea works in the wild.
What this is not
Honesty is the point, so this is stated plainly. AgentScore has no revenue and no paying customers. It is a solo-operated asset, not a staffed company. The detection surface is pattern-based, not a full static-analysis or behavioural engine. If you are looking for a book of business or a team to acquihire, this is not that. If you are looking for an MCP-security wedge, a genuinely unique scan dataset, a credible disclosure record, and a registry foothold that would take months to rebuild, it may be worth a conversation.
Likely buyers
Teams already extending into MCP or software supply-chain security, MCP registries and directories that want a per-server risk signal, and security-research groups that would value the longitudinal dataset and disclosure archive. The founder is not available for employment; this is structured as a clean asset purchase with a short paid transition if useful.
Make an approach
Acquisition inquiries go to security@agentscores.xyz. Tell us who you are and what you are building; we will share scan samples, dataset extracts, and the full asset inventory under a short conversation. Operated by Janus Compliance Limited (UK company number 16583861).