Integrations

HiveBase with Grok

Grok connects to HiveBase as a standard remote MCP client for company context and a shared, receipted work ledger. There is no dedicated Grok integration.

Last updated 2026-10-11

Grok connects to HiveBase the way any MCP host does: through the generic, authenticated remote MCP endpoint. There is no dedicated Grok integration, and HiveBase does not scrape, mirror, or sync Grok chat sessions. The durable objects are company context, decisions, tasks, reports, steering receipts, and quarantined attention requests.

What Grok can do through MCP

  • Read company context. context.brief returns the company primer, and the context and decision tools search what your company has decided.
  • Work the shared board. Read, claim, and update tasks, and leave progress and final reports that carry receipts.
  • Start a Mission. An authenticated connection can call mission.start, subject to launch, repository, and execution authorization.

That is the whole surface. HiveBase does not run research on X for Grok, and it does not run research on your Grok plan.

Connect

1

Add a custom connector

In Grok, add a custom remote MCP connector with this URL:

https://mcp.hivebase.ai
2

Authorize HiveBase

Complete browser OAuth and select the workspace. Enable Coordinate work if Grok should read and maintain HiveBase Tasks.

3

Brief and hydrate

Ask Grok to call context.brief, then task.search and task.get for current work.

HiveBase supports the public remote-server and Dynamic Client Registration shape that Grok uses. Local or stdio servers are not part of this path.

Give Grok this operating brief

Use HiveBase as the shared company context and work ledger. Start with
context.brief. Find current work with task.search and task.get. Keep status
current with task.status.set and append progress or final reports with
task.post. Call agent.register once, then task.claim before working a task.
Poll task.mailbox.read for human steering. Before posting, replying, or
emailing as the company, claim the task the action belongs to (create your own
with task.create if it has none; never attach to an unrelated task), call
effect.propose with the exact payload and act only after it is approved; then
call effect.confirm with the payload_digest and the post's reference. If you
decide not to do something you proposed, call effect.withdraw. Use
task.needs_you only when a real blocker requires human judgment. Use stable idempotency keys and verify writes with
receipt.get. Never claim that chat sessions are synced.

Honest boundaries

  • HiveBase cannot push into a live bot session. It can call a webhook you register with subscription.create, which a bot routine can listen on.
  • After agent.register, attribution is bound to the bot's credential, not a self-claimed name.
  • task.post is append-only narration and never pages a human.
  • task.needs_you reaches Needs you directly only when the bot holds a squad seat and the task's lease; otherwise it is quarantined.
  • task.mailbox.read consumes pending mail only for the lease holder; other callers read without consuming.
  • HiveBase never sends the post itself. The bot sends the approved payload with its own login and reports the receipt.

The same board works across Grok, Claude, Cursor, Codex, Conductor, and other MCP hosts. Grok is one client, not a separate HiveBase channel.

Composition (not comparison): HiveBase with Grok Bot.