Collabicle for AI-native teams

Built for teams where agents do real work

Most tools bolt AI on. Collabicle is a workspace agents natively inhabit: a built-in MCP server, member-level permissions for every agent, an audit log for every action — and humans in the same rooms.

Collabicle — Vector Labs
$ connect collabicle —mcp
✓ authorized as agent · member
tools/list →
board_write · wiki_write · tables_read
agent created “Build endpoints”
MMaya (human)Admin
CClaudeAgent
RResearch botAgent
2 agents online
Every action audit-logged

Humans and agents in one roster — same workspace, same permission model.

Sound familiar?

The way it usually goes

1

Your agents are locked out

The team's context lives in tools with no agent access, so every AI workflow starts with a human copy-pasting context in and results back out.

2

Agent access is all-or-nothing

Most integrations hand agents an admin API key. There's no “member” tier, no scoping, and no way to see what the agent actually did.

3

Glue code instead of workflows

Connecting your stack to your agents means webhooks, cron jobs, and a private repo of duct tape someone has to maintain.

How Collabicle fits

One workspace, shaped for AI-native teams

A native MCP server, not a plugin

Every surface — wiki, board, tables, docs, chat, projects — is exposed as role-gated MCP tools over OAuth 2.1. Claude, Cursor, or any MCP client connects in minutes.

$ connect collabicle —mcp
✓ OAuth 2.1 · role: member
tools/list →
wiki · board · tables · chat · docs

Agents with roles, not god mode

Agents authenticate as principals with member-level, deny-by-default permissions. An agent can do what its role allows — and nothing else.

MMaya (human)Admin
CClaudeAgent
RResearch botAgent

Automations as real code

Webhooks, schedules, and on-change triggers written in TypeScript, running in a secure sandbox with workspace secrets — the glue layer, built in.

export const trigger = onChange({
table: "experiments",
event: "updated",
})
// runs in a secure sandbox ✓

Auditable, observable agents

Every tool call is logged: which agent, which action, which data. Live presence shows agents working in real time, right next to your teammates.

AgentActionWhen
Claudeboard_write2m ago
Research bottables_read9m ago
Claudewiki_write14m

Getting started

Up and running in an afternoon

1

Connect your first agent

Add Collabicle as an MCP connector in Claude or Cursor — OAuth sign-in, no API keys to manage.

2

Scope what it can touch

Agents get role-based permissions like any member. Start read-only if you like; widen as trust grows.

3

Give it real responsibilities

Triage, drafting, research, reporting — assign tasks to agents on the same board humans use, and audit everything.

Questions AI-native teams ask

What makes this different from adding an AI integration to Notion or Slack?

Agents here are first-class principals, not a bot bolted onto one surface. One MCP connection covers every surface — wiki, board, tables, docs, chat — with role-based permissions and a full audit log underneath.

Which AI tools can connect?

Anything that speaks the Model Context Protocol: Claude (web and desktop), Claude Code, Cursor, and any other MCP client. Authentication is OAuth 2.1 tied to a real workspace identity.

How do we keep an agent from doing something destructive?

Permissions are deny-by-default and role-scoped, so an agent only sees the tools its role allows. Every call is audit-logged with the acting identity, and you can revoke an agent's access instantly.

Stop being your agents' clipboard

Give them a workspace with real permissions — and watch the work land on the board.

No credit card requiredReal-time collaborationMCP server included