Skip to main content
Integrations · Agent toolingOn the device

AI coding assistantsKnow what your assistants reach for.

Coding assistants call tools a staff member can wire up without asking anyone. RedactWall governs calls routed through its MCP guard and reports declarations received from enrolled workstations; bypass paths stay outside it.

In practice

What that looks like

An assistant routed through RedactWall reaches for a tool nobody approved

The call is refused, and you can see that it was attempted.

A tool hands the assistant a file full of customer data

Checked before the model reads it.

A developer wires up AI tooling on an enrolled workstation

It appears as a finding, not as approved configuration.

An enrolled workstation has not completed its declared check

Recorded as unchecked, so it cannot pass for clean.

What it takes to switch on

Configure the MCP client to start its local server through the RedactWall guard. Calls made through another route are not governed by this control.

How far this is proven

Runs on the workstation

The MCP guard runs in front of the local server it is configured to launch. Its proof does not extend to servers or tools the assistant reaches by another route.

Every limit, in one place: Disclosures.

Review the proposed first pilot

The proposed first pilot covers ChatGPT text in managed Chrome and a separately verified Anthropic gateway path. Other integrations are part of the broader product.

  • Coverage is published for this integration
  • Proof status is shown explicitly
  • Pilot scope is agreed before onboarding