MCP Server Cortex

What gbrigandi/mcp-server-cortex MCP server does

gbrigandi/mcp-server-cortex MCP server exposes selected Cortex analysis workflows as tools for MCP-compatible clients. Cortex performs the underlying observable analysis through enabled analyzers, while the MCP layer accepts tool parameters, starts jobs, waits for completion, and returns the resulting report when successful.

The supported inputs include IP addresses, domains, fully qualified domain names, URLs, and email addresses, depending on the tool.

How it works

The server connects to the Cortex API using an endpoint and an API key. A tool call selects an analyzer, either by using its default analyzer name or by supplying a configured analyzer instance. After starting a Cortex job, the server polls for completion up to the requested retry limit and returns the job report or a Cortex error.

The gbrigandi/mcp-server-cortex MCP server does not provide the external threat-intelligence credentials itself. Those credentials belong in the relevant analyzer configuration inside Cortex. For example, AbuseIPDB, VirusTotal, and urlscan.io analyzers may each require their own API key.

Setup and configuration

Use a precompiled binary from the repository's GitHub Releases page, place it in a suitable directory, and make it executable on Linux or macOS. The project also supports building from source with the Rust toolchain. An MCP client must be configured to launch the executable; the README gives Claude Desktop as an example.

Set these environment variables before starting the process:

• CORTEX_ENDPOINT: Full Cortex API URL, such as http://localhost:9000/api.
• CORTEX_API_KEY: Cortex API key with permission to list analyzers and run jobs.
• RUST_LOG: Optional logging configuration, including values such as info.

The Cortex instance must be reachable from the server, and the analyzers used by the tools must be enabled and configured in the relevant Cortex organization. Testing each analyzer in the Cortex interface first can help identify configuration or credential problems.

Tools and capabilities

The gbrigandi/mcp-server-cortex MCP server provides these documented tools:

• analyze_ip_with_abuseipdb accepts an IP address, an optional analyzer name defaulting to AbuseIPDB_1_0, and an optional retry limit defaulting to five.
• analyze_with_abusefinder accepts data and a type of ip, domain, fqdn, url, or mail. Its default analyzer is Abuse_Finder_3_0, and it also accepts a retry limit.
• scan_url_with_virustotal submits a URL to a VirusTotal analyzer, defaulting to VirusTotal_Scan_3_1, with an optional retry limit.
• analyze_url_with_urlscan_io submits a URL to a urlscan.io analyzer, defaulting to Urlscan_io_Scan_0_1_0.

Analyzer names can be overridden where the tool exposes that parameter, allowing a Cortex deployment to use a differently named analyzer instance.

Limitations and notes

Analysis depends on the Cortex deployment and its analyzer configuration. A disabled analyzer, missing provider API key, invalid credentials, or other Cortex-side configuration issue can cause a tool call to fail or return an error. The server also needs network access to the Cortex API.

Retry limits control polling attempts and do not replace Cortex job execution or provider-side rate limits.