KernelScan

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Linux kernel CVE analyzer: upload a .config, get a CycloneDX VEX report of affecting CVEs.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "KernelScan": {
      "url": "https://kernelscan.io/mcp/",
      "transport": "streamable-http"
    }
  }
}

Measured tools (11)

• search_cves

  Search Linux kernel CVEs. No API key required: keyless callers get the free public tier — recent high-severity Linux kernel CVEs (capped at 25 results). Free *keyed* callers see only CVEs published in the last 60 days; basic+ keyed callers get the full corpus. ``query`` matches against CVE id and description (case-insensitive). ``severity`` filters by effective severity (``critical``/``high``/``medium``/``low``). ``cvss_min`` filters by effective CVSS score. ``published_after`` (ISO 8601) returns only CVEs newer than that date. Returns up to ``limit`` (max 100) CVEs, newest first.

  Parameters:

  • limit — integer
  • query
  • cvss_min
  • severity
  • published_after

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• get_cve

  Fetch a single Linux kernel CVE by ID (e.g. ``CVE-2024-12345``). No API key required: keyless callers get the public representation of a CVE, but only for CVEs in the public set (recent high-severity); any other id returns ``not found``. Free *keyed* callers get a 404 for CVEs published more than 60 days ago. AI risk-summary / analysis fields are included for any keyed user on CVEs in the public set, and for pro / enterprise on every assessed CVE.

  Parameters:

  • cve_id (required) — string

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• list_products

  List the calling user's products with denormalized analysis stats. Paid plans only (basic / pro / enterprise). Free callers get a clear upgrade message.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• get_product

  Fetch one product owned by the caller, including the CVE breakdown. Returns 404 (not 403) if the product belongs to another user, so product existence isn't leaked across accounts.

  Parameters:

  • product_id (required) — string

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• get_product_vex

  Return the VEX MANIFEST for one of the caller's products — metadata, not the document. Multi-megabyte CycloneDX documents (up to 8,000+ vulnerability entries) are not safe model-context payloads, so this tool returns a bounded manifest: CycloneDX format/spec version, product id, generated/expires timestamps, the VEX hash and the composite ETag identity, the uncompressed size in bytes, total + per-status vulnerability counts, and the authenticated REST download path. Reads from the 24h ProductVexCache; if the cache is empty/expired the next call to ``get_product`` (or the REST endpoint) will regenerate it. The MANIFEST carries the same ``kernelscan.io:exploit_maturity`` / ``kernelscan.io:kev`` overlay identity as the REST download (backend#337), so ETags compare across transports. To inspect the entries themselves, use ``list_product_vex_entries``. To retrieve the COMPLETE CycloneDX document, use the authenticated REST endpoint ``GET /api/products/{product_id}/vex`` (same ks_live_ key) — that is the canonical way to retrieve the full artifact; no MCP tool returns it.

  Parameters:

  • product_id (required) — string

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• list_product_vex_entries

  Page through the VEX vulnerability entries of one of the caller's products. Returns COMPLETE CycloneDX vulnerability objects for one bounded page — never the root document, never all entries — plus ``returned``, ``total_matching``, and a ``next_cursor`` to continue with. Every result is bounded to 256 KiB: the page stops before the byte limit and returns a cursor when necessary. Entries are ordered by CVE id (ascending) and carry the same live ``kernelscan.io:exploit_maturity`` / ``kernelscan.io:kev`` properties as the REST download (backend#337). Filters (all optional, combinable): - ``statuses``: ``affected`` / ``not_affected`` / ``in_triage`` - ``severities``: ``critical`` / ``high`` / ``medium`` / ``low`` / ``none`` - ``kev``: true/false — CISA KEV listing only / non-KEV only - ``exploit_maturity``: ``poc`` / ``weaponized`` - ``cve_ids``: exact-match list of CVE ids ``limit`` defaults to 25, maximum 100. ``cursor`` is the opaque continuation token from a previous page — it is tied to the product, the active filters, AND the current document + threat-overlay revision: changing filters, a regenerated cache, or a KEV/PoC signal that moved since the last page invalidates it (start a fresh page without a cursor — re-using a stale one is rejected, never silently re-applied). Within one revision, concatenating all pages yields each matching CVE exactly once. The COMPLETE multi-megabyte CycloneDX document is served by the authenticated REST endpoint ``GET /api/products/{product_id}/vex`` (same ks_live_ key) — the canonical way to retrieve the full artifact.

  Parameters:

  • product_id (required) — string
  • kev
  • limit — integer
  • cursor
  • cve_ids
  • statuses
  • severities
  • exploit_maturity

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• create_product

  Create a new product, run analysis, and return its initial stats. ``config_upload_id`` references a previously-staged .config that the caller POSTed to ``/api/configs/uploads`` over plain HTTP — the LLM does NOT emit the config text itself (a real kernel .config is ~100–200 KB and exceeds a single tool-call output budget). Workflow: 1. Caller / wrapper script: ``curl -H "Authorization: Bearer ks_live_..." \ -F "[email protected]" \ https://kernelscan.io/api/configs/uploads`` returns ``{config_upload_id, sha256, size_bytes, expires_at}``. 2. Pass that ``config_upload_id`` into this tool. Uploads are per-user, single-use, and expire 30 minutes after upload. Same gates as POST /api/products: free can't create products; paid plans are capped at their resolved product limit — read it (and any per-account override) from ``whoami.product_limit`` rather than assuming a fixed per-tier number. ``factor_ids`` are silently ignored unless the plan allows security factors (``whoami.can_use_factors``). Re-using a product name returns 409. Creating a product RUNS an analysis, so it spends one unit of the team's SHARED monthly analysis allowance (``whoami.monthly_analyses_used`` / ``monthly_analyses_limit``). When the allowance is exhausted the tool fails with "Monthly analysis limit reached (…/month) [429]". This is a durable monthly quota — NOT the transient per-call rate limit that also surfaces as 429: it will not clear until next month, so report it to the user instead of retrying. Check ``whoami`` before a batch of creates. ``kernel_version`` is the kernel's release. For a stable kernel that is its version (``6.6.67``). For a CIP SLTS kernel pass its CIP release or the full ``uname -r``: ``4.19.325-cip136``, ``4.19.325-cip136-rt50``; a trailing local suffix such as ``-yocto-standard`` is accepted and ignored. The ``-cipN`` counter decides which of CIP's own backported fixes apply, so pass it whenever the device runs a CIP kernel — a bare ``4.19.325`` is analysed as the final 4.19 stable release (the ``.config`` never names the ``-cipN``). The known CIP releases of a series are listed by ``GET /api/kernel-releases?flavor=cip&series=4.19`` (add ``&rt=1`` for the RT tree). The returned ``kernel_release`` shows how the string was read (``flavor``, ``base``, ``series``, ``cip``, ``rt``, ``local``; null when the string is no release). Surrounding whitespace is trimmed. A version string the analysis cannot read (e.g. ``4.19.325cip136``, or anything non-ASCII) fails with "Unrecognized kernel version" [400] before anything is created or the upload is consumed. A CIP release must be on the stable base the uploaded ``.config``'s header names (the header shows the exact base; a CIP tag never changes it): ``4.19.300-cip90`` with a ``4.19.325`` config fails with "… is based on 4.19.300, but the .config header says 4.19.325" [400]. This is checked before the upload is consumed, so the same upload can be used again with the right release. A stable version, or a ``.config`` without a version header, is not checked.

  Parameters:

  • name (required) — string
  • kernel_version (required) — string
  • arch (required) — string
  • config_upload_id (required) — string
  • factor_ids
  • description

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• update_product

  Update a product owned by the caller. Re-runs analysis if the kernel_version, arch, or referenced .config changed. To change the .config, first POST the new file to ``/api/configs/uploads`` (see ``create_product`` for the curl recipe) and pass the returned ``config_upload_id`` here. Leave ``config_upload_id`` as ``None`` to keep the existing .config. ``factor_ids=None`` leaves factor selections untouched; an empty list clears them. Same tier gates as PUT /api/products/{id}. A change that re-runs analysis (``kernel_version``, ``arch``, or the ``.config``) spends one unit of the team's shared monthly analysis allowance and can fail with the same durable "Monthly analysis limit reached … [429]" quota error as ``create_product`` (distinct from the transient rate-limit 429 — don't retry it). A rename / description / factor-only edit runs no analysis and is free; the product's VEX is re-evaluated against the new factors in the background (not metered), so ``get_product`` / ``get_product_vex`` show the new verdicts a few seconds later. ``kernel_version`` takes the same forms as in ``create_product``: a stable version, or a CIP release / full ``uname -r`` such as ``4.19.325-cip136-rt50``, whose ``-cipN`` counter decides which CIP fixes apply. A changed version string the analysis cannot read fails with 400 and changes nothing; re-sending the stored value is always accepted. A CIP release must stay on the base its ``.config`` header names: whichever of the two the update changes is checked against the other ([400], before the upload is consumed).

  Parameters:

  • product_id (required) — string
  • arch
  • name
  • factor_ids
  • description
  • kernel_version
  • config_upload_id

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• whoami

  Return the caller's identity, plan, and quota state. Works without an API key: keyless callers get a lightweight public-tier payload (no account) describing how to request access.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• request_access

  Explain how to get a KernelScan account (no API key needed). Self-registration is open — there is no invitation to wait for and no admin in the loop. The user signs up on kernelscan.io themselves (email + password, accepting the terms), confirms the verification mail, and mints a ks_live_ API key on their account page. This tool only hands that path back: it files nothing and sends no mail. ``email`` / ``name`` / ``reason`` are still accepted so older clients don't break, but they are ignored — never tell the user that a request was submitted on their behalf.

  Parameters:

  • name
  • email
  • reason

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• submit_support_report

  Send a support / dispute report to KernelScan staff. Use this when an automated CVE or factor assessment looks wrong, or when you need to hand human-needed context back to the team. The caller's API-key user is attached automatically (id, email, plan) so support can look the account up. ``category`` should be one of: - ``cve_assessment`` — wrong AI verdict / CVSS / CWE on a CVE - ``factor_assessment`` — wrong factor verdict for a product - ``bug`` — broken behavior in the API or UI - ``other`` — anything else ``cve_id`` / ``product_id`` / ``assessment_id`` are optional but recommended — they let support jump straight to the relevant row.

  Parameters:

  • category (required) — string
  • subject (required) — string
  • message (required) — string
  • cve_id
  • product_id
  • assessment_id

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

Publisher website

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.