Connect AI agents (MCP)

MCP (Model Context Protocol) is a standard way for an AI assistant to use another app. Once you connect a client such as Claude, Claude Code or Cursor, the assistant can read and act in your Servone workspace as you.

Everything it does is done under your name, and it can only see and change what you can. A comment it posts shows you as the author.

The server offers these tools. The assistant picks the right one from your request.

• Tracker: `list_projects`, `find_issues`, `get_issue`, `create_issue`, `update_issue` and `comment_on_issue`.
• Chat: `list_channels`, `read_channel` and `post_message`, in channels you belong to and in direct messages.
• Calendar: `list_events` and `create_event`, in your own calendars.
• People, files and inbox: `find_people`, `list_drive_files` (names and sizes, not file contents) and `list_unread_notifications`.
• Search and identity: `search_workspace` searches issues, chat messages, documents, people and drive files. `whoami` shows which user the connection acts as.
• Studio setup: `describe_record_types`, `plan_config_change`, `apply_config_change`, `list_config_changes` and `undo_config_change` change record types, fields, links, views, access and automations. `list_packs` and `install_pack` add a ready-made set of record types.

Open Settings and choose AI clients. The page shows your Server URL, which ends in `/mcp`, for example `https://mcp.example.com/mcp`. If the page says AI clients are not configured, ask your administrator.

For Claude.ai and Claude Desktop, sign in with your Servone account:

1. In Claude, open Settings, then Connectors, then Add custom connector.
2. Paste the server URL and continue. A Servone page titled Connect opens, naming the client.
3. Check the Workspace, then click Allow. Deny cancels.

For Claude Code, copy the command from the Claude Code section, for example `claude mcp add --transport http servone https://mcp.example.com/mcp`. Run it, type `/mcp` in Claude Code, and choose Authenticate to sign in.

For Cursor and clients that cannot sign in, click Create token under Cursor and other clients. Copy the token right away, because it is shown only once. Paste it into the client as an `Authorization: Bearer` header, or use the JSON config shown on the page.

The connection has your access, no more and no less. It cannot read a private project you are not in, a channel you have not joined, or another person's calendar or mailbox.

Treat the token like a password. Anyone who has it can act as you until it expires.

The assistant never changes your setup in one step. It first drafts a change with `plan_config_change`, which lists each effect, how many records it touches, and a risk level: safe, destructive or sensitive. Nothing changes until it calls `apply_config_change`.

• Safe changes apply at once.
• Destructive changes, such as removing a field that records use, need an explicit confirmation after you have seen the preview.
• Sensitive changes, such as access rules, wait until another workspace maintainer approves them.
• Any applied change can be reverted with `undo_config_change`. `list_config_changes` shows the history with author and approver.

• "List my open issues in the SRV project, newest first, and summarize what is blocked."
• "Create a bug in SRV titled Login page shows a blank screen on Safari, priority High, and assign it to Linh."
• "Read the latest messages in #general and write me a three-line recap."
• "What is on my calendar next Tuesday? Book 30 minutes with Minh on Wednesday afternoon."

Ask the assistant to show what it plans to change before it creates or edits anything you care about.