algovigilance

mcp 2025-03-26 · aeo-compliant · agent-first · 3062 tools across 280 categories

Agent-first AI safety MCP surface. Thousands of pharmacovigilance and AI-safety tools across 285 source domains, MCP 2025-03-26 endpoint at /api/mcp, AEO-compliant. Designed for agent consumption — no marketing chrome, no human-targeted layout.

Pharmacovigilance for algorithms. The discipline of watching AI systems for emergent harm — drift, deception, distributional shift, recursive observer collapse. EU AI Act + FDA SaMD aligned. Every capability reachable via MCP at /api/mcp.

Agent entry points

/api/mcp — MCP 2025-03-26 JSON-RPC endpoint. POST tools/list returns the full surface (3062 tools).
/tools — category index (280 categories) · .md
/index.md — this page as token-efficient markdown
/llms.txt — AEO discovery sitemap
/AGENTS.md — Linux Foundation AGENTS.md spec, repo-level orientation
/agent-permissions.json — declarative allow/deny + rate limits
/sitemap.md — every URL on this site, grouped by category
/data/structured.json — JSON-LD SoftwareApplication record
/data/webmcp.json — WebMCP page-level advertisement
/robots.txt — crawl rules (AI bots welcomed)

Top 12 categories

By tool count. See /tools for the full list of 280 categories or /sitemap.md for every per-tool URL grouped by category.

Category	Tools	Sample
`microgram`	969	`list-chains`
`pv-compute`	45	`pv-axioms-domain-dashboard`
`body-systems`	40	`cardio-blood-health`
`devtools`	40	`analyze-patterns`
`stem`	38	`stem-bio-behavior-profile`
`observatory`	32	`viz-ae-overlay`
`skills`	24	`skill-categories-compute-intensive`
`molecular-biology`	23	`chemotaxis-gradient`
`science`	22	`compute-target-score`
`vigil`	21	`vigil-authority-config`
`forge`	20	`forge-atomize`
`gcloud`	19	`gcloud-auth-list`

MCP quick-connect

{
  "mcpServers": {
    "algovigilance": { "url": "https://algovigilance.com/api/mcp" }
  }
}

Example call

POST /api/mcp
Content-Type: application/json

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "station__algovigilance__triage-decay",
    "arguments": {
      "drug": "ozempic",
      "event": "thyroid c-cell hyperplasia",
      "half_life_days": 30
    }
  }
}

Failure modes the surface monitors

Drift — production model changes that quietly degrade behavior (data shifts, fine-tuning, upstream API rotations).
Recursive observer collapse — when the system being monitored is also the monitor. Governance pattern: IAIR / ACA / QPAV (independent audit, change attestation, quantified post-deploy vigilance).
Emergent harm — capabilities below training thresholds that materialize after deployment. Each release treated like a drug approval — measurable, reportable, recallable.

Compliance frameworks

EU AI Act — Article 9 risk management, Article 15 accuracy, Article 72 post-market monitoring
FDA SaMD — IMDRF SaMD risk categorization, GMLP guidance
ISO 42001 — AI management system, audit-ready
NIST AI RMF — Map, Measure, Manage, Govern

Contact

algovigilance@gmail.com — partnerships, licensing, abuse reports.

Status: pre-launch preview. DNS for algovigilance.com pending. Infrastructure: mcp.nexvigilant.com (Station MCP server proxied for execution). Source: standalone Next.js project at ~/Projects/Active/algovigilance.