> For the complete documentation index, see [llms.txt](https://docs.countercyclical.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.countercyclical.io/agents/governance.md).

# Governance

Manage workspace policies, approvals, model and tool availability, spending controls, and data access for Agents.

Navigate to [**Settings › Workspace › Agents › Governance**](https://dashboard.countercyclical.io/settings/workspace/agents/governance). Governance settings apply at the workspace level and determine the capabilities Agents can use when executing work.

See also [Approval & Access](https://docs.countercyclical.io/agents/approval-and-access) and [Custom MCP Servers](https://docs.countercyclical.io/developers/mcp-servers/custom-mcp-servers).

### Agent Access & MCP

**Agent Access & MCP** controls how Agents and external MCP clients can interact with the workspace.

* **External MCP Connectors** — Allows Agents to communicate with approved tools outside Countercyclical through MCP.
* **Countercyclical MCP Access** — Allows authenticated external MCP clients to access resources in the workspace.
* **Full Access** — Allows workspace owners and admins to run permitted Agent actions without routine human approval. Workspace security, compliance, and audit controls continue to apply.

The displayed **Revision** identifies the current revision of these governance settings.

#### Full Access and Approvals

Full Access changes how eligible governed actions are approved; it does not bypass the rest of the governance configuration. An Agent still cannot use a disabled model, disallowed tool, unavailable data source, or capability prohibited elsewhere by workspace policy.

Use Full Access when trusted owners and admins need less approval friction while retaining workspace-level controls.

### Model Allowlist

The **Model Allowlist** determines which AI providers and models Agents may use in the workspace.

Models are organized by provider. You can enable or disable an entire provider or control individual models beneath it. The summary at the top shows how many available models are currently enabled.

Disabling a model removes it as an available choice for governed Agent execution. Provider-level controls provide a faster way to enable or disable the provider's models together.

### Tool Allowlist

The **Tool Allowlist** defines which capabilities Agents may invoke. Available tools shown in the workspace include:

* **Search Web** — Searches the live web for current or externally verifiable evidence and returns source information suitable for citation.
* **Search Workspace** — Searches connected workspace accounts and data sources using permission-aware retrieval.
* **Search Files** — Searches permitted file-like content indexed from connected accounts and data sources.
* **Create Artifact** — Allows an Agent to draft an artifact such as a report, memo, dashboard, workflow, file, or other supported output.
* **External Action** — Allows an Agent to request an external side effect, such as sending data, updating another system, or invoking a third-party action.

Tools marked **Approval** represent governed capabilities for which execution can require human authorization under the applicable approval policy.

Disabling a tool removes that capability from Agent execution regardless of the selected model.

### Spend Caps

**Spend Caps** provide workspace-level budget controls for Agent activity.

Turn on **Spend cap enabled** to configure a **Soft budget limit** and **Hard budget limit**, specified in cents. The soft limit provides a budget threshold for monitoring, while the hard limit defines the maximum permitted budget under the configured policy.

**Enforce hard limit** determines whether the hard budget is enforced as an execution constraint. This provides a safeguard against Agent activity exceeding the workspace's permitted spend.

Budget controls should generally be configured alongside model availability: allowing higher-cost models while setting an appropriate hard limit provides flexibility without leaving aggregate Agent expenditure unconstrained.

### Data Source Controls

**Data Source Controls** determine which categories of information Agents may retrieve while working.

The workspace provides separate controls for:

* **Web Search** — Permits retrieval from the public web.
* **Enterprise Search** — Permits retrieval from approved connected enterprise sources.
* **File Search** — Permits retrieval from approved file-based sources.

Where applicable, administrators can further restrict retrieval using **Allowed Connected Accounts** and **Allowed Data Sources**.&#x20;

These restrictions provide a narrower boundary than simply enabling a search capability: a search mechanism may be available while only explicitly permitted accounts or sources are accessible.

### How Governance Controls Interactions

Agent execution is constrained by the intersection of these settings rather than by any single control. For an Agent to perform a governed operation, the required model, tool, data source, access mode, and budget capacity must all be permitted.

For example, enabling **Search Workspace** in the Tool Allowlist does not by itself grant unrestricted enterprise-data access. The corresponding data-source controls and underlying user or workspace permissions still determine what the Agent can retrieve. Likewise, enabling **External Action** makes that capability available, but approval requirements and Full Access policy determine whether an eligible action can proceed automatically.

This layered model lets administrators separately govern **what Agents can use, what information they can access, what actions they can take, how much they can spend, and when human approval is required**.

### Governance Visibility in the Composer View

Workspace governance is surfaced directly in the **Agents** experience so users can understand the policies that constrain an Agent before starting work.

The **Agent Governance** summary displays the current workspace-enforced configuration without requiring users to leave the Agent composer.

Select **Manage Agent Governance** to open the workspace's Agent governance settings, where authorized administrators can modify these controls.

The summary is informational: governance remains centrally enforced at the workspace level. Changing a model, effort level, approval mode, or other option in the Agent composer cannot override a restriction imposed by workspace governance.

#### Related Pages

{% content-ref url="/pages/LFvzJEH7Zasj1u5wVRnA" %}
[Approval & Access](https://docs.countercyclical.io/agents/approval-and-access.md)
{% endcontent-ref %}

{% content-ref url="/pages/VCcrSyerDeZDgFScChHr" %}
[Custom MCP Servers](https://docs.countercyclical.io/developers/mcp-servers/custom-mcp-servers.md)
{% endcontent-ref %}

{% content-ref url="/pages/nzVomWwXZ1mLSiD6Z6Cv" %}
[MCP Servers](https://docs.countercyclical.io/developers/mcp-servers.md)
{% endcontent-ref %}
