Endpoint Configuration Locations
This page shows where the HYPR endpoint client applies managed-tool configuration on a workstation. Each tool keeps its settings in a different place on each operating system. Some tools also have an enterprise location, where configuration cannot be overridden by the developer.
The endpoint client writes these files for you. Knowing the locations helps you plan a managed rollout, and answer a developer who asks what changed on their machine.
This page is background for administrators and IT teams. Managed workstations apply this configuration automatically, and nothing here requires manual editing on a developer's machine.
User and Enterprise Locations
Configuration is applied per operating system, and each tool supports up to two kinds of location:
- User locations are in the developer's home directory. A tool is configured there by default. The endpoint client merges into them, so the developer's other settings survive.
- Enterprise locations are system-level paths, deployed through your device-management (MDM) tooling. The tool gives configuration in an enterprise location precedence, and a developer cannot override it. This is how you enforce a setting fleet-wide.
The endpoint client applies each setting at the first location it can write. It prefers an enterprise location that exists on the workstation and is writable, and otherwise uses the user location. It applies one location, not both, and records which one it used.
How the Client Preserves a Developer's Settings
The client follows three rules when it changes a tool's configuration:
- Backup before the first change. Before it changes a tool's file for the first time, the client saves a copy of the original. The copy stays unchanged from then on, and is what Restore tools to pre-HYPR config puts back. A file that did not exist is recorded as absent, so restore removes it.
- Merge, never overwrite. Routing settings are written to a single key. MCP servers are added to the tool's existing server list. Cursor hooks are appended to each event, alongside the developer's own. A file the client cannot parse is left alone, and reported as a failure.
- Remove only what was added. A workstation stands down when managed mode is off, or management is disabled for it. The client then removes the settings, servers and hooks it added, and leaves everything else in place.
Per-Tool Locations by Operating System
Each tool's configuration lives in the following places. User paths use the developer's home directory, shown as ~. Enterprise paths are absolute, per operating system.
Claude Code
Routing is applied through the ANTHROPIC_BASE_URL setting:
- Without an enterprise settings file, the endpoint client sets it as an environment variable in the developer's login session. It reaches Claude Code sessions started after the change.
- With an enterprise
managed-settings.json, the client writes the setting into that file'senvblock instead, so a developer cannot override it.
MCP servers are written to the user-level Claude Code configuration, ~/.claude.json. The locations are:
| Location | Path |
|---|---|
| User (MCP servers) | ~/.claude.json |
| Enterprise (macOS) | /Library/Application Support/ClaudeCode/managed-settings.json |
| Enterprise (Linux) | /etc/claude-code/managed-settings.json |
| Enterprise (Windows) | C:\Program Files\ClaudeCode\managed-settings.json |
For Claude Code on Windows, the current enterprise location is C:\Program Files\ClaudeCode\. The older C:\ProgramData\ClaudeCode location is deprecated.
Codex
Codex is configured through ~/.codex/config.toml. Routing uses openai_base_url, and MCP servers use the mcp_servers section. The endpoint client edits the file in place, and preserves the developer's comments, ordering and other settings.
If the CODEX_HOME environment variable is set, it overrides the default ~/.codex location, and the client follows it. Codex has no separate system-wide location.
Cursor
Cursor is configured for MCP servers and for observation hooks. Its enterprise hooks location differs per operating system:
| Location | Path |
|---|---|
| User (MCP servers) | ~/.cursor/mcp.json |
| User (hooks) | ~/.cursor/hooks.json |
| Enterprise hooks (macOS) | /Library/Application Support/Cursor/hooks.json |
| Enterprise hooks (Linux) | /etc/cursor/hooks.json |
| Enterprise hooks (Windows) | C:\ProgramData\Cursor\hooks.json |
Cursor reads its hooks from every location it supports, and runs all matching hooks. Managed hooks are therefore merged into the hooks file, alongside a developer's own. The managed hooks report to the HYPR endpoint client only, and never block an action.
Enforcing Routing Fleet-Wide
Combine routing with Egress Enforcement, so a routed tool also cannot reach its provider directly.
If your organization builds its own endpoint client instead of deploying the HYPR endpoint client, contact your HYPR representative for the endpoint-client integration guide. It describes the configuration-profile and enforcement interfaces in detail.
Related
- Routing and Observation — how routing is applied to each tool
- Egress Enforcement — how direct traffic is blocked per tool
- Endpoint Platform Support — what each operating system's endpoint client applies