Claude Tag gets an admin API: 14 endpoints to control Claude in Slack

Anthropic has added a Claude Tag API to its Claude Platform release notes dated October 23, 2026. It lets Claude Enterprise organizations read and change, from code, the settings that decide where and how Claude answers in Slack.

The entry appeared on the release notes page, which lists changes to the Claude API, the client SDKs and the Claude Console. Anthropic’s documentation for the API gives most of the detail below.

Fourteen endpoints for channels, workspaces and bundles

The API is in beta for Claude Enterprise organizations and needs the beta header experimental-claude-tag-2026-10-21. According to Anthropic’s documentation, requests without the header return a 404. Its endpoints sit under /v1/organizations/claude_tag/. With them, an admin can inspect and modify Claude Tag configuration for individual Slack channels and for whole Slack workspaces, look up the effective settings in a channel together with the origin of each value, and link Access bundles to channels or workspaces.

The documentation counts 14 endpoints across four resources: five for Slack channels, three for workspaces, two read-only ones for bundles and four for bundle targets. Each channel, workspace and the organization’s Slack defaults can store five optional settings: an on/off switch, standing instructions (up to 8,192 characters), a model, a cloud environment and a guest policy with three modes (full access, restrict, channel only).

An Access bundle is a named set of connectors, repositories and credentials. The API can list bundles and attach them to a channel or a whole workspace, but it returns only a name, description and timestamps, never what a bundle grants. It cannot create or edit bundles.

Settings resolve through three levels. The nearest level wins for the on/off switch, model, environment and guest policy, while instructions from all levels are joined, broadest first. Adding include_inherited=true to a read request shows the settings that actually apply, with the source of each value.

Who can call it, and what it cannot do

Access requires an Admin API key, issued in claude.ai by the organization’s primary owner. A key carrying read:claude_tag can view configuration, while write:claude_tag allows changes; the write scope does not include reads. Anthropic says the API is not available to Claude Platform (Claude Console) organizations, and the documentation adds that it covers Slack only. The Python SDK exposes it under client.beta.organization.claude_tag, and the ant CLI has matching commands.

Several limits matter for anyone planning to use it. Removing a channel is permanent: it deletes settings, sessions, transcripts, memory, routines and artifacts, and is allowed only while the channel’s switch resolves to off. Connecting or disconnecting workspaces, editing the Slack defaults and changing spend limits are outside the API. If any value in a settings request is refused, nothing in that request is stored, and the last write wins because there are no preconditions.

Anthropic’s documentation suggests three uses: gating Claude on approved channels through an internal approval system, keeping configuration in a repository and applying only the differences, and auditing for risky states such as the tool being on without approval or guests given full access. The release note itself includes no pricing, and Anthropic has not said when the API will leave beta.

What to watch as Claude Tag spreads

Claude Tag itself is newer than the API. Reuters, as carried by The Star, reported that Anthropic launched it on June 23, 2026 for Claude Enterprise and Team customers. Users call Claude into a channel or thread with an @Claude mention, and administrators control the channels, tools and data it can reach. Anthropic said in that launch that Claude Tag replaces its earlier Claude in Slack app, and that 65% of its product team’s code is created by an internal version of the tool. That is Anthropic’s own claim and has not been independently checked.

The API suggests that, four months on, large customers are asking for management that does not depend on clicking through a settings page. An agent that can read a workspace’s conversations and use connected repositories is something security teams tend to want to inventory and gate. The audit workflow in the documentation points at that need.

Three signs would show whether the API is gaining traction. First, whether the beta reaches Team plans, since the API is Enterprise only while Claude Tag also runs for Team customers. Second, whether the gaps close: bundle creation and the Slack defaults are not manageable through the API today. Third, whether the experimental header is dropped, as Anthropic has done with other betas in the same release notes, such as the Files API and cache diagnostics. Until then, the interface can change.

The same page also lists other changes from the past week, including the October 21 release of API version 2026-10-21 with major new versions of the client SDKs and the launch of Claude Models on Google Cloud (Anthropic managed) on October 22. The Claude Tag API is a narrower, administrator-facing addition, but it is the first sign in these notes of Anthropic treating Claude in Slack as infrastructure that needs to be configured like any other.