Skip to main content
Cowork reaches a third-party endpoint through gateway mode. The gateway must implement the Anthropic Messages API, which we run at POST /v1/messages. Cowork gives you nowhere to set start_within, so the deadline goes on the key. Create one first (see agent keys). A duration races Claude’s cheaper tier on managed keys, without streaming, inside a window of three to ten minutes. An interactive session streams, so pick a tier here rather than Flex race. See Claude durations need a longer window. Your organization has to be able to run Anthropic before you start. On your own keys that means an Anthropic provider key on the Provider keys tab, because we don’t resell inference. On Managed Keys it means Anthropic turned on and funded. Cowork keeps these settings behind menus, so this page has no agent setup prompt. Use Export below to push the same configuration across a fleet.

​
Enable developer mode

The Developer menu stays hidden until you turn it on.
  1. Open Claude Desktop.
  2. Go to Help, then Troubleshooting.
  3. Press Enable Developer Mode.
Developer then appears in the menu bar.

​
Configure Cowork

  1. Go to Developer, then Configure Third-Party Inference….
  2. Set Inference provider to Gateway.
  3. Set Gateway base URL to the router, with no /v1 suffix.
  4. Paste your agent key into Gateway API key.
  5. Leave Credential kind on Static API key.
  6. Leave Gateway auth scheme on Bearer. We refuse x-api-key.
  7. Pick a claude-* model in the model picker.
Cowork reads our GET /v1/models response to fill the picker. It lists only models it recognizes as Claude, which is what you want here.

​
Deploy across a fleet

Click Export in the same window for a .mobileconfig on macOS or a .reg on Windows, then push it with your MDM. Three keys hold the integration. Set inferenceModels to an explicit list only if the picker misses a model you expect.

​
Leave the beta features off

Cowork turns Claude Code’s experimental beta features off by default, because strict gateways reject the anthropic-beta headers and request fields they add. We’re one of those gateways, and we don’t forward those headers. Leave toolSearchEnabled unset for the same reason. Turning it on sends beta headers we don’t forward, and requests then fail.

​
Confirm the key applied

Every response comes with x-flexinference-defaults-applied. Cowork doesn’t show response headers, so read the request in the dashboard under Logs instead.

​
Troubleshooting

There is no Developer menu. Developer mode is off, and it hides the menu entirely. Turn it on under Help, then Troubleshooting, then Enable Developer Mode. 401 missing_api_key. Gateway auth scheme is set to x-api-key, which we don’t read. Switch it to Bearer. The model picker is empty. Cowork filters GET /v1/models down to ids it recognizes as Claude. Confirm your organization can run claude-* models, or set inferenceModels yourself. 400 on every request after enabling tool search. toolSearchEnabled makes Cowork send experimental anthropic-beta headers and fields that we don’t forward. Leave it unset. 400 unsupported_parameter naming output_config. The picker selected a non-Claude model, and Cowork sends Anthropic-native fields we refuse to mistranslate. Pick a claude-* model. See errors for every refusal we return, agent keys for the ones that are not specific to Cowork, and Claude Code for the terminal client.