The workspace your AI agents can actually use
Most tools bolt AI onto one surface. Collabicle exposes the whole workspace — wiki, boards, tables, docs, chat — over the Model Context Protocol, with the same role-based permissions your team has. Connect an agent in minutes; audit everything it does.
Agents appear in the workspace roster like anyone else — with a role, a presence, and a paper trail.
How it works
From zero to a working agent in three steps
Connect over OAuth
Add Collabicle as a connector in Claude or Cursor and sign in. No API keys, no env vars — the agent gets a real workspace identity.
Tools arrive role-gated
The agent's tool list is filtered by its role, deny-by-default. Read-only agents see read tools; members see what members can do.
Work lands, logged
The agent reads specs, updates boards, writes tables, posts in chat — and every single tool call is recorded with its identity.
One connection, every surface
The whole workspace, exposed as tools
Not a chatbot in a sidebar — role-gated MCP tools over the same data your team works in, updating in real time.
Wiki & docs
Agents browse the knowledge tree by path, read specs, and draft pages.
wiki_read · wiki_write · documents_readBoard & projects
Create tasks, move statuses, respect dependencies, roll up milestones.
board_write · projects_readTables
Query records and write rows — typed fields and constraints keep data safe.
tables_read · tables_writeChat
Post updates and digests into channels, with mentions that link real work.
chat_read · chat_postTeams & members
Understand who's who, so summaries and assignments land correctly.
teams_read · list_workspace_membersAutomations
Inspect and trigger workspace automations as part of larger agent plans.
automations_read · automations_runSecurity
Agent access you can defend to your team
Giving agents write access to team data is a trust decision. The architecture is built so you can say yes carefully.
OAuth 2.1, not API keys
Agents sign in through the same auth system as people. Access is tied to a workspace identity you can see and revoke — no long-lived keys in config files.
Deny-by-default roles
The tool list itself is filtered by the caller's role tier. An agent that shouldn't write tables doesn't get a broken button — the tool simply doesn't exist for it.
Identity that can't be forged
The acting identity is injected by the gateway on the server side and stripped from client input. A tool call can never claim to be someone it isn't.
Every call audit-logged
Which agent, which tool, which workspace, when — every call lands in an audit log with per-workspace usage dashboards on top.
What teams hand their agents
Questions teams ask first
Which AI tools can connect to Collabicle?
Anything that speaks the Model Context Protocol: Claude on web and desktop, Claude Code, Cursor, and any other MCP client — including agents you build yourself. Authentication is OAuth 2.1 tied to a real workspace identity, so there are no API keys to copy around.
What exactly can an agent do once connected?
Whatever its role allows: read and write wiki pages, create and move board tasks, query and update table records, browse and import documents, post in chat, and manage projects. Tools are exposed per surface and filtered by the caller's role — an agent only sees what it's permitted to use.
How do you stop an agent from doing something it shouldn't?
Three layers. Permissions are deny-by-default and role-scoped, so the tool list itself is filtered. The caller's identity is injected server-side and can never be forged from the client. And every tool call is audit-logged with the acting identity, so there is always a trail — and access can be revoked instantly.
Is the MCP server an add-on or a paid tier?
Neither. The MCP server is part of every Collabicle workspace, including the free plan. Usage limits scale with plans, but agent access with role-based permissions is a core feature, not an upsell.
Put your first agent on the team today
Create a workspace, add the connector, and watch the work land on the board — audited, permissioned, real.