MCP server

The AlertaVuln MCP server lets any Model Context Protocol client - Claude Code, Claude Desktop, Cursor, or your own agent - query and manage your AlertaVuln organisation directly: list projects, triage alerts, pull SAST, DAST, and container findings, weigh reachability, check packages against CVE intelligence, and manage your tech stack and webhooks.

The hosted server is live at:

https://alertavuln.com/mcp

How it works

The server (internally mcp-router) is a small, stateless Go service that exposes the AlertaVuln REST API as MCP tools. It is a proxy: it holds no secrets and no database of its own. Every tool call is forwarded to the AlertaVuln API using the API key you present.

It supports two transports:

• HTTP (the hosted endpoint) - MCP streamable HTTP (JSON-RPC over HTTP), stateless. Your org API key is forwarded per request from the Authorization: Bearer av_live_... header (or X-AlertaVuln-Key if your client can’t set Authorization). The key is never stored - it lives only for the duration of that request’s tool calls.
• stdio (the local binary, default transport) - a single session where the key is read once at startup from the ALERTAVULN_API_KEY environment variable. ALERTAVULN_API_URL overrides the API base URL and defaults to https://alertavuln.com.

Get an API key

You authenticate with a per-organisation API key: the prefix av_live_ followed by 64 hex characters. Create one in the web app on the Settings page, in the API Keys panel (org-admin only). When you create a key you choose a name and a scope:

• ReadOnly - can call all read tools (list / get / check / find).
• ReadWrite - additionally allowed to call the write tools (create, update, delete, acknowledge, sync, test).

Create the key under Settings, in the API Keys panel - here named for the MCP server. Demo data.

Caution

The raw secret is shown exactly once, at creation time - it is stored only as a hash server-side and can never be retrieved again. Copy it immediately; if you lose it, revoke it and create a new one. Treat it like a password: it acts on behalf of your whole organisation.

Connect a client

There are two ways to run the server, and they need different setup:

• Hosted endpoint (https://alertavuln.com/mcp) - run by AlertaVuln. Nothing to install: point your client at the URL and present your API key. This is what the tabs below configure.
• Local mcp-router binary - runs on your own machine over stdio, for air-gapped setups or when you would rather not route your key through a bridge. You install it with the CLI; see Run the router locally below.

Replace av_live_YOUR_KEY_HERE with your key in each snippet.

Claude Code

Add the hosted server with one command:

claude mcp add --transport http alertavuln https://alertavuln.com/mcp \
  --header "Authorization: Bearer av_live_YOUR_KEY_HERE"

Or declare it in your project’s .mcp.json:

.mcp.json

{
  "mcpServers": {
    "alertavuln": {
      "type": "http",
      "url": "https://alertavuln.com/mcp",
      "headers": {
        "Authorization": "Bearer av_live_YOUR_KEY_HERE"
      }
    }
  }
}

Claude Desktop

Claude Desktop’s mcpServers config speaks stdio natively, so bridge to the hosted HTTP endpoint with mcp-remote. Add this to claude_desktop_config.json and restart Claude Desktop:

claude_desktop_config.json

{
  "mcpServers": {
    "alertavuln": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://alertavuln.com/mcp",
        "--header",
        "Authorization: Bearer av_live_YOUR_KEY_HERE"
      ]
    }
  }
}

If your Claude Desktop build supports remote MCP servers natively, point it directly at https://alertavuln.com/mcp with the same Authorization header instead.

Cursor

Add the server to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global):

.cursor/mcp.json

{
  "mcpServers": {
    "alertavuln": {
      "url": "https://alertavuln.com/mcp",
      "headers": {
        "Authorization": "Bearer av_live_YOUR_KEY_HERE"
      }
    }
  }
}

Any MCP client

Any client that supports the streamable HTTP transport can connect directly:

• Endpoint: https://alertavuln.com/mcp
• Header: Authorization: Bearer av_live_... (or X-AlertaVuln-Key: av_live_...)
• Protocol: MCP JSON-RPC 2.0 over stateless streamable HTTP

Run the router locally (stdio)

For air-gapped setups, or when you prefer not to send your key through a bridge, the same binary runs as a local stdio MCP server. Supply the key via the environment instead of a header:

{
  "mcpServers": {
    "alertavuln": {
      "command": "mcp-router",
      "env": {
        "ALERTAVULN_API_KEY": "av_live_YOUR_KEY_HERE",
        "ALERTAVULN_API_URL": "https://alertavuln.com"
      }
    }
  }
}

In stdio mode the key is read once at startup; ALERTAVULN_API_URL may be omitted - it defaults to https://alertavuln.com.

Installing the local binary

The mcp-router binary ships with current CLI releases. Install it alongside the CLI with --with-mcp-router (install.sh) or -WithMcpRouter (install.ps1), or download mcp-router-<os>-<arch> from the release store and verify it against SHA256SUMS. See Install the CLI. The hosted endpoint above needs no install at all.

Available tools

The server registers 41 tools. Tools marked write require a ReadWrite-scoped key; a read-only key calling them gets back “this API key is read-only or lacks permission”.

Organisation

Tool	Access	Description
get_org	read	Get your organisation (id, name, contact email, timestamps)
list_members	read	List members of your organisation
list_invites	read	List pending / recently accepted invites
get_audit_log	read	One page of the org audit log

Projects

Tool	Access	Description
list_projects	read	List all projects in your org
get_project	read	Get a single project by id
create_project	write	Create a new project
update_project	write	Update a project’s settings
delete_project	write	Delete a project and all its data

Alerts

Tool	Access	Description
list_alerts	read	List a project’s vulnerability alerts, triaged REDYELLOWGREEN; filter by status, acknowledged, resolved, page
get_alert	read	Get a single alert by id within a project
acknowledge_alert	write	Acknowledge an alert, with an optional note

SAST (code scanning)

Tool	Access	Description
list_sast_findings	read	List a project’s current SAST findings, tiered RED / YELLOW / GREEN; filter by state (open / all / fixed), severity, and source repo
list_sast_scan_jobs	read	List a project’s server-side scan-job history (status, source repo, tier counts, queued time)

DAST (running-app scanning)

Tool	Access	Description
list_dast_findings	read	List a project’s current DAST findings (a dynamic scan of a running web app or API - injection, XSS, misconfiguration, exposures), tiered RED / YELLOW / GREEN; filter by state (open / all / fixed), severity, and target URL
list_dast_scan_jobs	read	List a project’s DAST scan history: each run’s mode, target, RED / YELLOW / GREEN counts, and start time
list_dast_targets	read	List a project’s registered DAST scan targets (URL, name, environment, last-scanned)
list_dast_server_scan_jobs	read	List a project’s hosted DAST scan-job history: each job’s status, target URL, tier counts, and queued time

Reachability

Tool	Access	Description
list_reachability_verdicts	read	List a project’s reachability verdicts: per (CVE, package) whether the vulnerability is likely exploitable here, with an effective-priority score, the usage / exploit / exposure evidence behind it, and any human override; optionally filter to one CVE
get_reachability_verdict	read	Get a single reachability verdict by id within a project
override_reachability_verdict	write	Record a human override on a verdict (NotExploitable or Exploitable) with a required reason; overrides win over recomputation

Overriding a verdict is more restricted than a normal write tool: it requires a Business or Enterprise subscription, and the override endpoint accepts member (Bearer JWT) sessions only, so an org API key cannot call it - through the MCP server it fails with a 401.

Container images

Tool	Access	Description
list_container_images	read	List a project’s scanned container images: each image’s ref, digest, OS, whether continuous monitoring is on, and its latest RED / YELLOW / GREEN counts
list_container_findings	read	List a project’s container image findings (vulnerable OS packages and dependencies baked into image layers), tiered RED / YELLOW / GREEN; filter by image, state (open / all), and severity. Each finding notes whether it comes from a base-image layer or one of your own
get_base_image_advice	read	Get base-image upgrade advice for one scanned image: alternative base images with the RED / YELLOW / GREEN count delta each would bring, plus the Dockerfile FROM line to change when the project has a connected repo

Tech stack

Tool	Access	Description
list_tech_stack	read	List a project’s tracked packages with health flags
add_tech_stack_item	write	Add a package to a project’s tech stack
update_tech_stack_item	write	Update a tech-stack item in place
remove_tech_stack_item	write	Remove a package from the tech stack
sync_tech_stack	write	Reconcile a project’s tech stack from a parsed dependency manifest

Webhooks

Tool	Access	Description
list_webhooks	read	List a project’s notification webhooks
add_webhook	write	Add a notification webhook (slack, googlechat, teams, discord, generic)
update_webhook	write	Update a webhook
delete_webhook	write	Delete a webhook
test_webhook	write	Send a test notification to a webhook

Intelligence

Tool	Access	Description
check_package	read	Pre-flight a single package coordinate against AlertaVuln intelligence: matching CVEs (severity / CVSS / EPSS / KEV), maintenance health, suggested safe version
find_exposed	read	Find every project in your org with an alert for a given CVE id (blast-radius lookup)
get_package_health	read	List a project’s at-risk (deprecated / unmaintained) packages
export_sbom	read	Export a project’s SBOM (CycloneDX by default, optionally with VEX) - requires an Enterprise subscription

Aggregate queries

These read-only tools compose the lists above into bounded, decision-shaped answers - the questions an assistant is actually asked. Output is always a fixed top-N plus counts, never a full row dump, so the payload size is the same for a project with thousands of findings as for one with a handful.

Tool	Access	Description
get_findings_diff	read	Summarise what changed over a window (default 7 days) per domain - dependencies, code, running-app, containers - as new / resolved / open counts plus up to five of the worst new findings
get_top_risks	read	Rank the most urgent findings across domains into one list: known-exploited (KEV) alerts first, then findings reachable in your code, then remaining critical items by exploit probability; de-escalated and triaged findings are reported as counts, not ranked
get_posture_summary	read	A project’s current posture: open counts by domain and severity, acknowledged / suppressed totals, the reachability-verdict distribution, and a one-line 7-day trend

Verify the connection

The hosted server exposes an unauthenticated liveness endpoint:

curl -i https://alertavuln.com/mcp/health
# expect: HTTP/1.1 200 OK ... ok

With auth in place, an MCP initialize handshake identifies the server as alertavuln (version 0.1.0), and tools/list enumerates the tools above. If your key is missing, malformed, or revoked, tool calls fail with “authentication failed” (HTTP 401 from the API).

See also

• CI/CD - gate pipelines with the CLI
• av sast scan - the local scan whose findings list_sast_findings surfaces
• Install the CLI - the installer that also delivers mcp-router