
Internal MCP registry Your list, read on every call
Argent Strategic Intelligence keeps one registry of the MCP servers in your tenant. The MCP Gate reads it on every call, and every tool on it is fingerprinted, held for approval and re-checked on a schedule.
One list of the servers your agents may reach
The internal MCP registry is Argent's own list of the MCP servers in your tenant. It is the live inventory the MCP Gate governs: if a server is not on it, the gateway refuses the call.
Every tenant has it from day one. Servers get on the list three ways: an admin registers one by hand, a sync imports one from a registry you connect, or discovery finds one and an admin approves it.
One source of truth for what agents can reach, read by the gate on every call.
What the registry holds
| Part | What it is |
|---|---|
| Server record | Transport, address or launch command, the vault reference for its credential, sign-in style (none, bearer or a custom header), pinning mode, on or off, publisher and summary, and an icon. |
| Local servers | For servers Argent starts itself, an allowlist of the environment variables they may see. |
| Argument rules | The per-server rules that narrow a permission to certain arguments. |
| Tools | Every tool the server offers, each with a fingerprint of its definition and a state: pending, approved, denied, drifted or removed. |
| History | Each tool's past definitions, scan results and a diff against the approved version, so a change is reviewed before it is approved again. |
Argent re-reads every registered server on a schedule (by default every 15 minutes, and every 5 minutes for servers with pinning on Enforce), so a changed tool is caught without anyone pressing a button.
Bring your own registry
Many enterprises keep their own approved list of MCP servers. The MCP Registry's maintainers expect exactly that: "Private subregistries will exist within enterprises that have strict privacy and security requirements" (MCP blog).
Argent connects to a registry of your own that speaks the same API as the public one (the External MCP Registry connector). Give it the base address. Its servers import switched off and pending approval, exactly like the public sources.
Public sources work the same way: the Official MCP Registry and the Docker MCP Catalog.
Publish only what you approved
Argent can serve your approved servers back out, in the MCP Registry's own format, so MCP clients discover only the servers you allow.
- Same API shape. Argent serves the registry's read endpoints from this deployment, listing approved servers only.
- Capability tokens. Access to that list needs a registry token. An admin mints one on the MCP Connect page, sees it once, and can list or revoke it at any time.
- Client settings to match. The same page generates copyable settings for Claude Code and VS Code that keep each client on the gateway.
What you configure
There is nothing to install: the internal registry is part of every tenant.
- Look at it under Integrations, MCP registries. Test reads the live list for your tenant and reports how many servers it holds and their names.
- Add servers by hand, or connect a registry and sync. Approve what you want.
- Approve tools as they arrive under Needs a decision.
- Review discovered servers. Servers that discovery finds wait on the Discovery page for an approve or deny.
- Mint a registry token if your MCP clients should discover servers from Argent.
Where this fits
- Integration: MCP servers, the gate that reads this registry on every call.
- Sources that feed it: the Official MCP Registry, the Docker MCP Catalog, and your own registry.
- Capabilities: agent reach control, shadow AI discovery (MCP tools agents use that are not on the list), watch before you enforce.
- Goals: see all AI in your company, threat defense, sovereign AI (the registry works the same in an air-gapped on-prem install).
- Use cases: controlling agent tool access, AI in regulated and air-gapped environments.
Browse all integrations.
Want this running on your own use cases?
Book a demo