Claude
Connect your organization's Claude surfaces to Airia — telemetry from Claude Code and Cowork, browser coverage for claude.ai, and optional Anthropic usage and compliance data.
The Claude connection has two parts:
- Secure the Claude surfaces. Pick which Claude services your organization uses — Claude Code, Cowork, claude.ai — and Airia gives you the per-service setup to route their traffic through Airia, so usage is visible and guardrails apply.
- Optionally pull Anthropic usage and compliance data into your inventory for post-runtime audit. This needs an Anthropic organization admin or compliance API key, typically on Claude Enterprise.
This page covers the Discover → Connect side: the Claude setup page, what lands in your inventory, and how to verify it. For the full four-surface build — gateway routing, guardrails, agent constraints, and MDM rollout — see Securing Claude with Airia.
At a glance
| Surface | Cloud AI |
| What it does | Secures your Claude surfaces by routing them through Airia, and can pull usage and compliance data |
| How Airia connects | A setup page: pick the Claude services to secure, plus an Airia API key. For Claude.ai, an Anthropic compliance access key |
| Who sets it up | An Airia admin; an Anthropic org admin only for the optional compliance-data path |
| Time | ~15 min |
Before you start
- Know which Claude surfaces your organization uses: Claude Code, Cowork (Claude Desktop), claude.ai in the browser.
- Permission in Airia to create an API key. The setup page can create one for you.
- Optional, for the compliance-data path: an Anthropic organization admin or compliance API key with read access, on a plan that exposes this data.
Use an organization-level admin or compliance key scoped for reading, never a personal key. Treat it as a secret — never paste an API key into email, chat, or a ticket. This connection is read-only; it does not change anything in your Claude account.
Setup
Part A — The Claude setup page
Open the Claude setup page
In Airia, go to Discover → Connect and click the Claude tile.
Pick the services to secure
Under Services to secure, check the Claude surfaces your organization uses: Claude Code, Cowork, Claude.ai. Each one you check gets its own tab below with its configuration.
Create the Airia API key
Under Airia API key, click Create new key, or select an existing key. Claude Code and Cowork use this key.
Keys are show-once. Create the key here to fill it into the Claude Code and Cowork configuration; Copy JSON config and the download buttons unlock once a new key exists.
Configure Claude Code
On the Claude Code tab, deploy the settings shown to each user's ~/.claude/settings.json.
Use Copy JSON config, Download macOS script, or Download Windows script. The macOS
script covers Claude Code in VS Code too. Push the settings fleet-wide with your device
management tool.
Every Claude Code session then streams its telemetry (user prompts, tool calls, and raw API bodies, tagged with the user's email) to your Airia tenant.
Configure Cowork
On the Cowork tab, follow the steps shown. In https://claude.ai/admin-settings/cowork,
open Monitoring, set the OTLP endpoint, Headers, and Protocol to the values on
the tab, and click Save. Cowork includes each employee's email automatically.
Configure Claude.ai
On the Claude.ai tab, connect an Anthropic compliance access key. See Part C for the steps.
Confirm
Click Confirm to save the connection.
Part B — Roll out to each Claude surface
Two complementary mechanisms cover each surface: the telemetry feed for deep visibility, and gateway routing for inline control.
| Claude surface | Visibility | Inline control |
|---|---|---|
| Claude Code (terminal, VS Code) | The telemetry config from Part A — prompts, tool calls, API bodies stream to Airia | Point ANTHROPIC_BASE_URL at your Airia AI Gateway endpoint |
| Claude desktop app (Chat, Cowork, Code) | Cowork telemetry from Part A | Developer Mode → Configure third-party inference → Airia AI Gateway base URL. All three surfaces route; enable Allow Chat to include everyday chat. |
| Claude for Office add-ins | Not in Anthropic's audit logs — the gateway is the only visibility | Choose the LLM Gateway backend at Microsoft 365 deploy time |
| claude.ai web | Airia Browser Extension reporting | Airia Browser Extension — submit-time DLP, block/warn/redirect (no base-URL option exists) |
Why both matter. The telemetry feed gives near-real-time, per-user visibility including prompt content and tool calls, without needing an Anthropic Enterprise plan — but it observes, it can't block. Gateway routing applies DLP, prompt-injection defense, guardrails, and model routing before the prompt reaches Claude. That's the enforcement layer.
Verify it worked: send a throwaway prompt from one routed surface, for example Claude Code, and confirm the request appears in Airia's gateway usage and Discover dashboards within a few minutes.
Part C — Anthropic usage and compliance data
For post-runtime audit data pulled from Anthropic itself, connect a compliance access key on the Claude.ai tab. You need to be the Primary Owner of the Claude.ai organization, with the Compliance API enabled for your organization.
In Claude.ai, open your organization settings and go to Data Management Settings → Compliance access keys.
Click Create key and name it, for example airia-compliance. Select the scopes
read:compliance_activities, read:compliance_user_data, and read:compliance_org_data, then
click Create. Copy the key; it isn't shown again.
In Airia, on the Claude.ai tab, select an existing compliance credential, or click Create new, enter a Credential name and the Anthropic compliance key, and click Save credential.
Click Test Connection, then Confirm.
Have a user send a throwaway prompt in Claude, then confirm the activity appears in the Discover dashboards after the next refresh.
This path needs an Anthropic organization plan that exposes compliance data, typically Claude Enterprise. Everything it shows has already happened. It's an audit trail; real-time protection comes from the routing in Parts A and B.
Troubleshooting
| Symptom | Fix |
|---|---|
| Connection test fails | Re-check that the API key was copied in full, and that it's an organization admin or compliance key with read access — not a personal key. |
| Connected, but no data | Confirm the key has read access to usage and compliance data, and that your organization has Claude activity to report. Wait one refresh cycle. |
| A field you don't recognize | Field names vary by Airia build. Confirm what each field expects in Discover → Connect → Claude, or check with the Airia support team. |
| Key stopped working | The key may have been rotated or revoked in Anthropic. Create a new admin/read key and update the connection. |
What you get
| Data | How | Notes |
|---|---|---|
| Routed Claude traffic (Claude Code, Cowork, API) | The setup page plus AI Gateway routing | Real-time visibility and inline controls on every routed request |
| claude.ai browser usage | The Airia Browser Extension | Submit-time DLP, block/warn/redirect |
| Anthropic usage and compliance data (optional) | Compliance API polling, ~5 min | Post-runtime audit trail; needs Claude Enterprise |
It shows up in the Discover dashboards and inventory. The connection polls frequently — roughly every five minutes — so new activity appears after the next refresh.
What you can do
Discover — see Claude usage across the surfaces you route in real time and, optionally, from Anthropic's compliance data.
Govern and audit — usage lands in your inventory and dashboards. The optional compliance path adds a post-runtime audit trail for surfacing violations after the fact.
Secure — inline, real-time controls come from routing through the Airia AI Gateway (Claude Code and API via base URL, Desktop and Cowork via third-party inference) and the Airia Browser Extension for claude.ai.
Know which mechanism does what. The telemetry feed is deep monitoring — it observes, it can't block. Gateway routing and the Browser Extension are enforcement. The compliance data in Part C is after-the-fact audit.
| Capability | Supported |
|---|---|
| Discover | Yes |
| Govern | Yes, plus post-runtime compliance audit with Claude Enterprise |
| Secure | Yes, via AI Gateway routing and the Browser Extension; the setup page adds deep telemetry |
Related
- Securing Claude with Airia — the full four-surface build.
- Browser Extension — coverage for claude.ai in the browser.