Agents & MCP

MCP server · Streamable HTTP · read-only · no login
Your agent already has a list of CVEs. Zeptric tells it which ones matter today.

Scanners such as npm audit, pip-audit, Trivy, Grype and Dependabot produce the ids. What they cannot say is which of those are exploited in the wild, in CISA KEV, zero-days or used by ransomware, and so which to fix before lunch. That is what Zeptric answers, over the Model Context Protocol, from the same data that ranks this site.

https://zeptric.io/mcp

Connect

Any client that speaks MCP over Streamable HTTP

Claude Code

claude mcp add --transport http zeptric https://zeptric.io/mcp

Claude.ai and the desktop app

Settings → Connectors → Add custom connector. Name it Zeptric, paste https://zeptric.io/mcp as the URL and leave authentication empty.

Cursor, Windsurf, VS Code and other clients with a JSON config

{ "mcpServers": { "zeptric": { "url": "https://zeptric.io/mcp" } } }

Anything else

Every POST to the endpoint is one JSON-RPC request and gets one JSON answer; there is no session header to carry. The example below is a complete call.

Live example

6 ids from a JavaScript lockfile: 1 fix now, 2 fix soon, 3 to schedule. Answered by the API when this page was rendered.

IdPriorityWhy
CVE-2025-55182Meta React Server Components Remote Code Execution VulnerabilityFix now
In CISA KEVExploited in the wildUsed by ransomwarePublic exploitEPSS ≥ 50%CVSS 9+
CVE-2025-29927Authorization Bypass in Next.js MiddlewareFix soon
Public exploitEPSS ≥ 50%CVSS 9+
CVE-2025-7783Usage of unsafe random function in form-data for choosing boundaryFix soon
CVSS 9+
CVE-2024-4068Memory Exhaustion in bracesScheduleno exploitation signal yet
CVE-2024-21538Versions of the package cross-spawn before 7.0.5 are vulnerable to Regular Expression Denial of Service (ReDoS) due to improper input sanitization.Scheduleno exploitation signal yet
CVE-2024-55565nanoid (aka Nano ID) before 5.0.9 mishandles non-integer valuesScheduleno exploitation signal yet

The same call as one JSON-RPC request. No session, no token:

curl -s https://zeptric.io/mcp -H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"triage_vulnerabilities","arguments":{"ids":["CVE-2025-55182","CVE-2025-29927","CVE-2025-7783","CVE-2024-4068","CVE-2024-21538","CVE-2024-55565"]}}}'

The answer carries structuredContent with the items, the counts and the rule, so a model that sees only the output knows what the buckets mean.

Tools

Eight reads, all idempotent

ToolTakes, and answers with
triage_vulnerabilitiesup to 500 CVE, GHSA or ZDI ids
→ the ids ranked into fix now, fix soon, schedule and unknown, each with its signals, Zeptric severity, CVSS, EPSS, the KEV due date and known fixed versions
lookup_vulnerabilityone id
→ the full record: affected products and packages with fixed versions, exploits, vendor advisories, briefing, recent events and news
searcha vendor, product, actor, sector or topic name
→ matching tags with their slug and activity; an id gets a hint to use lookup_vulnerability
whats_newtag slugs, since (a date, RFC 3339, or 24h / 7d), event types, a severity floor
→ events newest first, with unknown_tags when a slug does not exist
search_breachesa company name or domain
→ breach claims and confirmed breaches, newest first
lookup_breacha breach id
→ sources, actor, sector, data classes, briefing, timeline and coverage
weekly_reportan ISO week like 2026-W46, or nothing for the current one
→ counts per event type, KEV additions, new exploits, top items
patch_tuesday_reporta month like 2026-11, or nothing for the latest
→ that day's vulnerabilities with their signals, exploited first

One prompt, triage_audit, takes an audit tool's output pasted as is and asks the model to extract the ids, call triage_vulnerabilities once and answer with fix-now, fix-soon and schedule lists.

The priority rule

Returned with every triage, so the reader of the output sees it too

  • Fix now in CISA KEV, exploited in the wild, a zero-day, or used by ransomware.
  • Fix soon a public exploit, EPSS of 10% or more, or CVSS 9 and above.
  • Schedule known, with no exploitation signal yet.
  • Unknown not in Zeptric. Never read it as safe.

Within a priority the higher Zeptric severity comes first. Severity is Zeptric's own 0 to 10 score, the one that ranks the feed; CVSS and EPSS are reported beside it, not instead of it.

REST

The same triage without MCP

curl -s https://zeptric.io/v1/vulns/triage -H 'Content-Type: application/json' -d '{"ids":["CVE-2024-3400","GHSA-xxxx-xxxx-xxxx"]}'

The spec is at /v1/openapi.yaml; any OpenAPI-aware agent framework can use it directly. Single lookups are GET /v1/vulns/{id}.

Limits and scope

  • 500 ids per triage call; 100 events per whats_new page, with a cursor for the next one.
  • The rate limit is the same per-address budget as the API, 120 requests a minute.
  • Nothing is stored. The server has no accounts, and a list of ids is answered and forgotten.
  • Zeptric does not scan. Bring the ids from your scanner; a tool that takes package names and versions directly is next on the list, and package-level coverage for npm, PyPI, Go, crates and Maven is still being filled in. An id Zeptric does not know comes back as unknown.

For crawlers and models there is /llms.txt; a GET on the endpoint returns a small card with the tool names and this page.