Connect your AI assistant
Connect your assistant to mdedit.ai to find, create, edit, review, export, and publish documents. Browser sign-in is the recommended setup for remote MCP clients that support it.
Connect with browser sign-in
- Install the mdedit Agent Plugin (opens in a new tab), or add
https://mcp.mdedit.ai/mcpin your assistant's remote MCP settings. - Use your assistant's sign-in option and sign in to mdedit.ai in the browser.
- Review workspace access and approve the connection.
All available workspaces are selected by default. You can deselect workspaces or choose read-only access before connecting. Your assistant can act only within the access you approve and your own workspace permissions.
The plugin includes installation options for VS Code, Gemini CLI, Cursor, and Claude Code. Other clients can use the server URL if they support remote MCP. Installation and sign-in controls vary by client; see the installation guide or the Claude Code guide.
Your assistant manages its saved connection. An older installation may ask you to sign in again after updating. Updating the plugin does not revoke your existing connections.
Use your CLI sign-in with a local MCP server
If your client launches local MCP processes, install the current CLI and sign in:
npm install -g mdedit-cli@latest
mdedit auth login
claude mcp add --transport stdio mdedit -- npx -y @mdedit/mcp-server@latestThe local MCP server shares the CLI's operating-system credential storage. Set MDEDIT_PROFILE=work to use a named profile. See CLI authentication.
API keys for automation
For clients without browser sign-in, CI, or unattended workflows, create a scoped key in Settings → Integrations → API Keys. Use a separate key for each integration and choose the permissions it needs. See API keys.
For a hosted connection, configure this URL and header in your client's private settings:
https://mcp.mdedit.ai/mcp
X-API-Key: YOUR_API_KEYFor a local process, supply the key through MDEDIT_API_KEY using your client's private environment settings, then run npx -y @mdedit/mcp-server@latest.
Keep credentials out of agent chat and shared project files.
Permissions
Your assistant may ask for additional permission when a task needs more access. Reconnect and approve the access you want it to have.
For API keys or custom integrations, common permission sets include:
- Review documents:
articles:read,reviews:write - Edit documents:
articles:write - Read publication state:
publishing:read - Publish or unpublish:
publishing:write - Convert files:
conversions:execute
Write permissions include the corresponding read operation. Publishing a document publicly still requires your explicit confirmation.
Public-link modes
publish_article always requires confirmPublic: true because it changes public
visibility. For a new link, omitting mode creates a Live link: later durable edits
update the same URL. Pass mode: "snapshot" for a frozen artifact.
For an existing link, omit mode to preserve its stored setting, or pass "live" /
"snapshot" to switch deliberately. get_publish_status returns the stable URL,
mode, source revision/version, last successful sync, and sync status. A failed Live
sync leaves the last good public content available.
Existing links created before Live mode are Snapshots until explicitly changed.
Manage or revoke access
Remove an assistant's workspace access in Team → Agent access to block new private document operations. Disconnect the account using your assistant's connection settings as well. Revoke API keys separately in mdedit.ai settings.
A previously issued download link remains usable until its expiry.
Troubleshooting
- Sign-in does not open: use your client's MCP authentication command and update the client if needed. For clients without browser sign-in, use an API key.
- Sign-in or connection errors: reconnect through your client's settings. Avoid deleting saved credentials unless the client's recovery instructions require it.
- Missing permissions: reconnect and approve the permissions required for the task.
- No workspaces or access denied: check that you approved the workspace and still have access to it in mdedit.ai.
- Unattended workflows: configure a scoped API key in private environment settings.
If the problem continues, contact support@mdedit.ai with your client name and version and the error message. Do not include credentials.
Last reviewed