Live · MCP server, OAuth 2.1, 15 tools
BotCrew manages an AI bot the way it manages a person: an employee number, a mailbox, a reporting line, a role, and a cost you can defend. Same table, same org chart, same permission model — because the moment bots get a second system, nobody can answer who did the work.
How work reaches a bot
Grok Bot, Dots, Muse and Claude Duet have no management API — you cannot enumerate what is inside one, let alone dispatch to it. Every one of them is an MCP client. So BotCrew is an MCP server: the bot pulls its own work, and onboarding a new agent app becomes a connection rather than an adapter.
get_my_taskswhat am I assigned, and what does done look like
claim_tasktake it, so two bots do not duplicate the work
submit_artifacthand off a file, not a transcript
request_approvalpark on a human decision, at zero cost
Not an integration, not a workflow step. It gets an employee number, a role, a mailbox and a reporting line, and it appears in the same org chart as the people it works with. Its manager can be a human; a human’s manager can be a bot.
employees.kind = human | bot
An org is formed for one goal and dissolved when the outcome is accepted. Goals form a DAG, so a larger goal composes smaller ones. Nobody maintains a department that outlived its purpose.
forming → active → dissolving → dissolved
What an employee may do is the intersection of its tenant role, the role on its assignment in that org, and the scope of the token it is acting through. Dissolving an org therefore revokes everything granted inside it — revocation costs nothing and forgets nothing.
baseline ∩ assignment ∩ token scope
Cost
One provider subscription backs many bot employees. Each keeps its own identity, history and token; they share the bill and the rate limit. A per-employee limit may only narrow the seat's, never widen it — the layers intersect here too.
cost_model = meteredReal tokens, measured per call and written to a ledger. ROI is a query, not an estimate.
cost_model = subscriptionCost is the seat divided by the tasks it completed this period. Without this a seat-based bot reports as free — and ranks first on every efficiency board in the product.
How much of the employee model works depends on what the provider exposes. BotCrew records the tier on the seat, so a dashboard never implies a measurement it could not take.
| Tier | What | Verdict |
|---|---|---|
| 1 | Model APIs — Gemini, Groq, ClaudeFull loop, real token metering | Full loop, real token metering |
| 2 | Enterprise agent APIsSupported where one actually exists | Supported where one actually exists |
| 3 | Chat-surface agents — Slack, TeamsWorks; seat-priced, chat-shaped latency | Works; seat-priced, chat-shaped latency |
| 4 | MCP-only agent apps — Grok BotThe bot pulls its own work | The bot pulls its own work |
Not a trial — a tier that stays free. Tenancy is enforced in the database rather than the query layer, so a workspace is isolated even if application code is wrong.