Custom MCP Servers

Custom MCP servers are privately developed packages (Python or npm) stored securely in JFrog Artifactory. They act as the bridge between AI agents and your company's internal systems, such as databases, internal APIs, and private platforms. By registering them in the JFrog MCP Registry, you bring proprietary tools under enterprise governance and make them discoverable to AI and coding agents (such as Claude Code, Cursor, and VS Code) through the governed JFrog Agent Guard.

Registering these custom MCP servers in the JFrog MCP Registry, enables:

Secure Discovery - AI and coding agents can find your internal tools.
Enforced Governance - Every action is vetted against enterprise security policies.
Comprehensive Auditability - Track which tools are used, when, and by whom.
Granular Access Control - Only authorized users can register and use specific servers.

Registration Flow

To ensure a trusted supply chain, the package and its optional README.md must first exist in a local Artifactory repository before it can be registered and governed by the MCP Registry.

Choose Your Custom MCP Server Type

JFrog supports two primary categories of custom MCP servers. Use the table below to determine the appropriate registration flow for your environment.

If your MCP server is...	Use
A Python or npm package uploaded to a local Artifactory repository	Register Local Custom MCP Servers
An external HTTPS endpoint hosted outside Artifactory (for example, an internal service)	Register Remote Custom MCP Servers

Implementation Use Cases

Type	Use Case
Local (Artifactory-hosted)	Use this flow when your team builds and ships the MCP server as a binary or package artifact. These servers are: Stored in local Artifactory repositories (NPM, Python, etc.). Scanned by JFrog security tooling for vulnerabilities and license compliance. Distributed through the same trusted supply chain used for standard software packages.
Remote (Endpoint)	Use this flow when the MCP server is hosted as a remote web service (for example, an internal microservice or a managed SaaS endpoint) and you only need to register its address. Infrastructure: The server is not hosted as an artifact but as a URL endpoint. Communication: Supports protocols like Server-Sent Events (SSE) or Streamable HTTP. Governance: The JFrog Agent Guard brokers and governs the connection on behalf of developers, enforcing policies defined in the AI Catalog.

Type

Use Case

Local (Artifactory-hosted)

Use this flow when your team builds and ships the MCP server as a binary or package artifact. These servers are:

Stored in local Artifactory repositories (NPM, Python, etc.).
Scanned by JFrog security tooling for vulnerabilities and license compliance.
Distributed through the same trusted supply chain used for standard software packages.

Remote (Endpoint)

Use this flow when the MCP server is hosted as a remote web service (for example, an internal microservice or a managed SaaS endpoint) and you only need to register its address.

Infrastructure: The server is not hosted as an artifact but as a URL endpoint.
Communication: Supports protocols like Server-Sent Events (SSE) or Streamable HTTP.
Governance: The JFrog Agent Guard brokers and governs the connection on behalf of developers, enforcing policies defined in the AI Catalog.

📘
Note on Docker Support:
Docker-based MCP servers are currently unavailable through Discovery (the public catalog) and manual registration as a "Custom MCP".

Next Steps

Verify you have the required permissions for the target project in the AI Catalog.
Ensure your environment has the necessary runtimes (Node.js, Python, or Docker) to support the gateway.
Navigate to AI/ML > Registry to register your internal services.