MCP access

MCP (Model Context Protocol) lets an external AI assistant your company already uses read CORA's governed company context, without anyone copying it by hand.

Read-only today

CORA's MCP server only reads. It cannot change anything in CORA, start work, or act in connected tools. If MCP actions arrive later, they will create the same CORA tasks as actions started in the app, with the same approvals and audit.

Tools

Each token lists the tools it may use. The available tools are:

• Search company memory: reviewed company facts, workflows, values, and decisions that are safe for all workspace members.
• Get company profile: the workspace profile and the shared Identity, Offerings, and Audience documents.
• Get company instructions: active, manager-approved operating rules.
• Search knowledge documents and Get a knowledge document: shared knowledge documents.
• List shared work items and Get a shared work item: work deliberately pinned to company knowledge. Private drafts and unpinned deliverables are left out.
• Search decisions: decision instructions and reviewed memories.
• Search meetings: words in recorded meetings' titles, notes, and what was said, as short passages with the speaker and meeting time.
• List recent meetings: recorded meetings, newest first, with participants and the memo summary.
• Read a meeting's memo and notes: one meeting's memo (summary, decisions, facts and context, open questions), its proposed follow-up, shared task suggestions, and what people wrote in the note themselves.
• Read a meeting transcript: what was said in one meeting, with speaker names and times, one page at a time.

Meetings and privacy commands

An assistant reads a meeting the way CORA does, a little more strictly. Speech covered by a privacy command, such as pause notes, excluding a topic, or redacting the last minutes, is never returned or searched. A few seconds of speech right next to a pause can be left out too, because the final transcript's times can differ slightly from the moment you gave the command. Lines that start with a voice command to CORA are left out as well. The transcript tells the assistant how many lines were left out, never what they said. Answers to questions asked to CORA during the meeting stay with the person who asked.

A token made before these meeting tools existed keeps the tools it was made with. To use them, create a new token with the meeting tools selected.

Permissions

A call through MCP reaches only what any ordinary workspace member could see. It never gets more than a member, whoever created the token.

Only workspace owners and admins can open MCP access, create tokens, and revoke them. Find it under Settings -> MCP access, or from the MCP access button in the Connections library.

Tokens and the call log

An assistant connects with an access token that belongs to one workspace, is limited to the tools you choose, can expire, and can be revoked at any time. Every call is recorded under Recent calls. See MCP tokens and revoking for the details, and Connect an AI assistant over MCP for setup.