> ## Documentation Index
> Fetch the complete documentation index at: https://docs.corridor.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Tool Controls

> Allow or block MCP servers and individual tools across your team's AI coding agents.

MCP Tool Controls add per-tool policy to Corridor's [MCP Compliance](/features/mcp-compliance). The policy page lists each MCP server observed across your team, with **Inherit / Allow / Block** controls for the server and every tool it exposes.

<Note>
  MCP Tool Controls are part of [Agent Governance](/agent-governance/overview) and are available on the **Enterprise** plan. Enforcement supports **Cursor**, **Claude Code**, **Factory Droid**, **OpenAI Codex**, and **Devin Desktop** through Hooks.
</Note>

## The policy page

Open **Governance → MCP** in the Corridor dashboard.

The page shows one row for each MCP server found in your team's [Artifact Inventory](/agent-governance/artifact-inventory). Each row shows the server's installations, observed tools, rules, and current policy. Expand the row to see its tools and installations, then select an installation to open its details.

Choose one setting for each server:

| Setting     | Behavior                                                                                |
| ----------- | --------------------------------------------------------------------------------------- |
| **Inherit** | The server follows your team default. Newly discovered servers start with this setting. |
| **Allow**   | Tools on the server are allowed unless a tool-level rule blocks them.                   |
| **Block**   | Tools on the server are blocked unless a tool-level rule allows them.                   |

Each tool has the same **Inherit / Allow / Block** settings. The page shows the tool's effective decision and whether it came from a tool rule, server setting, or team default.

The team-wide default at the top of the page, **Allow all** or **Block all**, applies to every tool without an explicit rule. Before you confirm a change, Corridor shows how many tools will switch.

### How Corridor resolves a decision

The most specific explicit setting wins:

1. **Tool rule**: An explicit Allow or Block on a tool overrides a server-level setting. You can block a server while allowing one approved tool, or allow a server while blocking a specific tool.
2. **Server setting**: If the tool has no rule, the server's Allow or Block setting applies.
3. **Team default**: If the server is set to Inherit, the team-wide default applies.

Corridor's MCP tools are always allowed and appear locked, so your policy cannot block Corridor's security analysis.

<Tip>
  A rule for a server also applies when a coding-agent plugin provides that server. You do not need a separate rule for the plugin-provided version.
</Tip>

### Reviewing new servers and tools

Newly discovered servers and tools show a **NEW** badge until someone marks them reviewed. The page also counts items that still need review. Corridor shows how each observed tool was discovered: from the server's published tool list, a session transcript, or a live agent tool call.

### Roles

| Action                                                                      | Who can do it                  |
| --------------------------------------------------------------------------- | ------------------------------ |
| Set server and tool policy, change the team default, or mark items reviewed | Team **Owners**                |
| View the policy page and inventory                                          | Team **Owners** and **Admins** |

## How enforcement works

When a developer's agent attempts an MCP tool call, Corridor's hook checks the call against your team's policy before it executes. A blocked call never reaches the MCP server. The agent receives a message explaining that your organization's compliance policy blocked the tool and that the developer should contact a team administrator.

Enforcement is **best-effort and fail-open**. If Corridor cannot check the policy, such as when the developer's machine is offline, the call is allowed so development can continue. [Agent Telemetry](/agent-governance/telemetry) records which tools were called, including calls that enforcement could not check.

<Note>
  Some locally launched MCP servers on Cursor and Devin Desktop are identified by their launch command. Configure these servers through the legacy MCP Compliance allow/block lists. The server and per-tool settings on this page do not apply to those calls.
</Note>

## Next steps

<CardGroup cols={2}>
  <Card title="Artifact Inventory" icon="boxes-stacked" href="/agent-governance/artifact-inventory">
    See every MCP server, skill, and plugin across your team
  </Card>

  <Card title="Shell Command Controls" icon="terminal" href="/agent-governance/shell-command-controls">
    Deny risky shell commands with bans and patterns
  </Card>
</CardGroup>