Deploy Coding Agent Governance with MDM

Coding Agent Governance configurations connect Cursor, Claude Code, Codex, and GitHub Copilot events to Endor Labs. Use this page to generate hook configuration for individual machines or managed fleets. You can create Cursor hooks.json, Claude Code settings.json, Codex requirements.toml, a GitHub Copilot policy hook file, or a macOS configuration profile for Claude Code or Codex. The generated hooks install and update endorctl when a coding agent session starts.

Before you begin

Complete the following tasks before you generate a configuration:

Complete the Coding Agent Governance prerequisites.
Create an API key with the AI Audit User role.
Identify the namespace that owns your Coding Agent Governance policies.
For manual generation, install git on the admin machine. The scripts run with system tools only. Creating a macOS configuration profile also requires macOS and plutil.

Generate a configuration in your browser

The browser generator assembles the configuration locally. Your API key, secret, namespace, and environment variable values are never sent to Endor Labs or GitHub. Use the manual generation steps below if the browser generator isn’t available or can’t fetch its source files.

Generate a configuration manually

The Endor Labs MDM scripts repository is the authoritative source for the generator scripts. Clone the repository on an admin machine.

git clone https://github.com/endorlabs/mdm-scripts
cd mdm-scripts/agent-governance

Set your credentials as environment variables. The generator reads these exact variable names, and prompts for any missing credential when you run it interactively.

export ENDOR_API_CREDENTIALS_KEY="<api-key>"
export ENDOR_API_CREDENTIALS_SECRET="<api-secret>"
export ENDOR_NAMESPACE="<namespace>"

Generate the configuration that matches your coding agent and deployment method.

Generate hooks.json for macOS or Linux. Cursor has no managed environment block, so every hook command carries the credentials and --env settings inline. Treat the file as a secret.

scripts/render.sh --agent cursor \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o cursor-hooks.json

For Windows, add --target-os windows. The generated hook uses an encoded PowerShell command that can run from Git Bash, PowerShell, or Command Prompt.

scripts/render.sh --agent cursor --target-os windows \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o cursor-hooks.json

Generate settings.json for a user installation, Linux managed settings, or Windows deployment.

scripts/render.sh --agent claude \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o claude-settings.json

For Windows, add --target-os windows.

scripts/render.sh --agent claude --target-os windows \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o claude-settings.json

Pipe the Claude Code settings through render-plist.sh to create a .mobileconfig file. Replace the identifier and organization with values for your organization.

scripts/render.sh --agent claude \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o - \
| scripts/render-plist.sh \
  --identifier com.example.ai-governance.claudecode \
  --organization "Example Organization" \
  --name "Claude Code - Endor AI Governance" \
  -o com.anthropic.claudecode.mobileconfig

Generate requirements.toml with managed hooks for macOS or Linux. Policy trusts hooks from a managed requirements source, so they run without per-user approval. Codex has no managed environment block, so every hook command carries the credentials and --env settings inline. Treat the file as a secret.

scripts/render.sh --agent codex \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o codex-requirements.toml

For Windows, add --target-os windows.

scripts/render.sh --agent codex --target-os windows \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o codex-requirements.toml

Pipe the Codex requirements through render-plist.sh --style mcx to create a .mobileconfig file that delivers the TOML as a managed preference. Replace the identifier and organization with values for your organization.

scripts/render.sh --agent codex \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o - \
| scripts/render-plist.sh --style mcx \
  --identifier com.example.ai-governance.codex \
  --organization "Example Organization" \
  --name "Codex - Endor AI Governance" \
  -o com.openai.codex.mobileconfig

Generate the policy hook file for macOS or Linux. Copilot has no managed environment block, so every hook command carries the credentials and --env settings inline. Treat the file as a secret.

scripts/render.sh --agent copilot \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o copilot-hooks.json

For Windows, add --target-os windows. The Windows session hook gets a 30-minute timeout because the bootstrap downloads endorctl during session start.

scripts/render.sh --agent copilot --target-os windows \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  -o copilot-hooks.json

The generated configuration contains your API key and secret. Treat the output and its MDM policy as secrets. Restrict access and rotate the API key if someone exposes the configuration.

Configure optional behavior

The generator enables the local response cache by default. Add --env ENDOR_AI_AUDIT_NO_BLOCKING=true to evaluate and record actions without blocking them.

scripts/render.sh --agent cursor \
  --api-key "$ENDOR_API_CREDENTIALS_KEY" \
  --api-secret "$ENDOR_API_CREDENTIALS_SECRET" \
  --namespace "$ENDOR_NAMESPACE" \
  --env ENDOR_AI_AUDIT_NO_BLOCKING=true \
  -o cursor-hooks.json

Add --skip-endorctl-update if your fleet provisions endorctl separately and shouldn’t check for updates when a session starts.

Deploy the generated configuration

Deliver the generated file through your mobile device management (MDM) or configuration management tool.

Deploy Cursor hooks

Deliver cursor-hooks.json to the location that matches your deployment:

For a local trial, copy cursor-hooks.json to ~/.cursor/hooks.json.
On managed macOS machines, deliver it to /Library/Application Support/Cursor/hooks.json.
On managed Linux machines, deliver it to /etc/cursor/hooks.json.
On Windows, push cursor-hooks.json to C:\ProgramData\Cursor\hooks.json with your MDM tool.

See Deploy hooks for Cursor to verify the installation and review the registered hook events.

Deploy Claude Code settings

Deliver the generated Claude Code configuration to the location that matches your deployment:

For a local trial, copy claude-settings.json to ~/.claude/settings.json.
On managed Linux machines, deliver it to /etc/claude-code/managed-settings.json.
On Windows, push claude-settings.json to C:\Program Files\ClaudeCode\managed-settings.json with your MDM tool.
On managed macOS machines, upload com.anthropic.claudecode.mobileconfig as a custom profile.

See Deploy hooks for Claude Code to verify the installation and review the registered hook events.

Deploy Codex requirements

Deliver the generated Codex configuration to the location that matches your deployment:

On managed macOS and Linux machines, deliver codex-requirements.toml to /etc/codex/requirements.toml.
On managed macOS machines, you can instead upload com.openai.codex.mobileconfig as a custom profile.
On Windows, push codex-requirements.toml to %ProgramData%\OpenAI\Codex\requirements.toml with your MDM tool.

See Deploy hooks for Codex to verify the installation and review the registered hook events.

Deploy Copilot policy hooks

Deliver copilot-hooks.json to Copilot’s policy directory. Copilot treats that directory as administrator policy: hooks there ignore disableAllHooks and run regardless of folder trust. Only the Copilot CLI reads the policy level.

For a local trial, copy copilot-hooks.json to ~/.copilot/hooks/endor.json. Both the Copilot CLI and VS Code agent mode read that directory, but a developer can edit it.
On managed macOS and Linux machines, deliver it to /etc/github-copilot/policy.d/endor.json, owned by root with mode 0644.
On Windows, push it to C:\ProgramData\GitHub\Copilot\policy.d\endor.json with your MDM tool.

See Deploy hooks for GitHub Copilot to verify the installation and review the registered hook events.

Update a deployment

The session hook installs endorctl on first use and checks for a newer version on later session starts. It verifies every downloaded binary with SHA-256. On macOS and Linux the hook never waits for the download. It hands the install or update to a background job and audits with the binary already on disk. A machine without endorctl skips the audit. The version check runs at most once a day. Set ENDORCTL_UPDATE_TTL_MINUTES in the configuration’s environment variables to change that interval. On Windows the hook downloads during session start. Regenerate and redeploy the configuration when you change credentials, environment variables, or delivery settings. Server-side Coding Agent Governance policy changes don’t require a new configuration.

Keep a fleet current with the scheduled runner

The mdm-scripts repository also provides scripts/runner.sh, which replaces one-shot generation with a self-updating deployment on macOS and Linux. Paste it into your MDM as a recurring script body: a Jamf script, a Kandji Custom Script, or a JumpCloud Command. Edit the settings block at the top to pick the agent (cursor, claude, codex, or copilot) and pin a reviewed revision. On each run it fetches the repository at that revision, renders the configuration on the endpoint, and replaces the installed file atomically, only when the result changed. Repository, credential, and flag changes then roll out on the next scheduled run without a redeploy. Keep the following in mind:

The MDM must set ENDOR_API_CREDENTIALS_KEY, ENDOR_API_CREDENTIALS_SECRET, and ENDOR_NAMESPACE in the script’s environment. Use an API key with the AI Audit User role.
Endpoints need only curl and tar, which come with macOS and Linux.
For copilot, the runner writes /etc/github-copilot/policy.d/endor.json and creates the directory if needed.
Windows does not use the runner. Pre-generate the configuration and push it with Intune or your MDM tool.

Introduction

Set up

Operate

Before you begin

Generate a configuration in your browser

Generate a configuration manually

Configure optional behavior

Deploy the generated configuration

Deploy Cursor hooks

Deploy Claude Code settings

Deploy Codex requirements

Deploy Copilot policy hooks

Update a deployment

Keep a fleet current with the scheduled runner

​Before you begin

​Generate a configuration in your browser

​Generate a configuration manually

​Configure optional behavior

​Deploy the generated configuration

​Deploy Cursor hooks

​Deploy Claude Code settings

​Deploy Codex requirements

​Deploy Copilot policy hooks

​Update a deployment

​Keep a fleet current with the scheduled runner

Before you begin

Generate a configuration in your browser

Generate a configuration manually

Configure optional behavior

Deploy the generated configuration

Deploy Cursor hooks

Deploy Claude Code settings

Deploy Codex requirements

Deploy Copilot policy hooks

Update a deployment

Keep a fleet current with the scheduled runner