Grodmir — MCP proxy for tool data protection

Grodmir is an MCP proxy between agents and connected MCP servers: the customer gateway checks tool arguments toward servers and results toward the model under signed policies.

Grodmir is an AI gateway for data protection and AI usage governance in organisations.

What does an MCP proxy inspect?

Grodmir is an MCP proxy between agents and connected MCP servers: the customer gateway checks tool arguments toward servers and results toward the model under signed policies.

Link to this answer

How do tool permissions and pins work?

Policies can block tools, require confirmation and pin approved tool definitions by hash. New or changed definitions stay hidden until reviewed and pinned; pinning supports review of tool poisoning and later definition changes, without proving that a tool is safe.

Link to this answer

What happens to pseudonyms in tool results?

Pseudonyms in MCP tool results are not restored across later tool use. Sending a placeholder introduced into model context back to a server is refused, to avoid writing placeholders over real values.

Link to this answer

When should an organisation evaluate Grodmir MCP proxy?

Evaluate Grodmir when agents use MCP servers and you need inspection of tool arguments and results, approved tool definitions and signed tool permissions. Connect each intended server through the proxy or central gateway and test client support for confirmations before rollout.

Link to this answer

Should you use a local MCP proxy or a central MCP gateway?

The local proxy wraps a configured MCP server connection on a workstation or exposes a loopback endpoint for a remote server. The central gateway gives an organisation an HTTPS endpoint per remote MCP server with OIDC sign-in, group-based tool rules and user-bound audit metadata. Both use the customer inspection boundary. The central gateway is self-hosted with a single SQLite replica; real commercial identity providers and vendor MCP servers still require deployment-specific verification.

Link to this answer

Does MCP tool pinning prove that a tool is safe?

No. Pinning checks that a tool definition matches the reviewed hash; changed or unapproved definitions are hidden and their calls refused. It does not prove that the approved definition or server implementation is safe, or that every malicious instruction will be detected. Combine definition review with tool permissions, supported confirmations, inspection of arguments and results, and controls that prevent direct unproxied connections.

Link to this answer

How can you verify this behaviour?

Content updated:

Protection covers connected-channel traffic only; detection can miss data and raise false alarms. Grodmir — AI gateway protection limits

Categories: AI gateway · Prompt DLP · MCP proxy · AI usage governance

Organisation portal