
Docker MCP Catalog Curated servers, your rules
Argent Strategic Intelligence reads the Docker MCP Catalog and imports each server switched off, so your admins approve what agents may use and the MCP Gate governs every call.
Docker's curated MCP servers, under your rules
Docker describes its MCP Catalog as "a curated collection of verified MCP servers, packaged as Docker images and distributed through Docker Hub" (Docker docs).
Connect it to Argent and those servers become candidates your admins can approve one by one. None of them is trusted, enabled or granted a tool until someone decides it should be.
Docker curates the catalog. You still decide what your agents may call.
What Argent does with the Docker MCP Catalog
| Step | What happens |
|---|---|
| Read | Argent fetches the catalog file in one read (the public v2 catalog.yaml), with a size cap. No account is needed. |
| Your own file | Point it at a catalog .yaml or .json file of your own and Argent reads that file instead. |
| Import | Each entry becomes a server record in your tenant, switched off and pending approval, with its description and tool names filled in. |
| Shape | An entry with a remote address comes in as an HTTP server. An entry packaged as an image comes in as a local server that keeps the catalog's image reference on its record and needs its launch command before it can run. |
| Repeat | Syncing again adds new entries only and never changes a server you already approved. |
Nothing imported is trusted
A server imported from the Docker MCP Catalog arrives switched off. The gateway will not broker a single call to it until an admin approves it.
- Approve or deny. The MCP registries page lists every imported server waiting for a decision. Approve enables it and lets the admin pick exactly which of its tools are allowed; Deny keeps it off.
- Every tool measured. Once a server is approved, Argent runs the MCP handshake, reads its full tool list and stores a fingerprint of each tool's definition. A tool whose definition later changes is marked as drifted, and the change is audited.
- Poisoned descriptions flagged. A built-in scanner checks each tool description for injection phrases, path rules and look-alike characters, and reports what it found without copying the text.
- Then the normal gate. An approved tool still needs a permission per agent on the MCP Gate page. See MCP servers.
How it connects
Argent reads the Docker MCP Catalog from its own control plane, outbound only, through the same guarded connection every Argent integration uses.
- One address. Catalog URL: the catalog file to read. Leave it at Docker's public catalog or use your own file.
- Safe outbound connection. The fetch uses Argent's shared outbound connection, which requires modern encryption and refuses internal addresses, so a registry address cannot be used to reach your internal systems.
- Honest failures. A fetch that fails reports a short reason and imports nothing. It never shows a registry as reachable when it is not.
- Per tenant. Every imported server is stamped with your tenant; tenants never see each other's imports.
What you configure
An admin connects the Docker MCP Catalog in the Argent console under Integrations, MCP registries.
- Add the connector and give it the address. Test checks the settings; turn on the live read option to have Test fetch the catalog too.
- Sync. One click fetches the catalog and imports each new server as pending approval. The result reads, for example, "Imported 12 (pending approval)".
- Approve what you want. Approve a server and choose which of its tools are allowed; deny the rest.
- Grant and watch. Give agents permission on the MCP Gate page, start the MCP Gate in watch mode, then switch it to Enforce.
Syncing again later adds only new servers. A server you already approved or curated is left exactly as you set it.
Where this fits
- Integration: MCP servers, the gate every imported server ends up behind.
- Other sources of servers: Official MCP Registry, the internal MCP registry.
- Capabilities: agent reach control, prompt injection defense, shadow AI discovery, watch before you enforce.
- Goals: threat defense, see all AI in your company.
- Use case: controlling agent tool access.
Browse all integrations.
Want this running on your own use cases?
Book a demo