AI agents (MCP)

Available on Pro and Agency. EkhaneMark runs an MCP server at https://ekhanemark.com/api/mcp. Your coding agent connects with a personal token and works on your clients' feedback the way you would: it reads the pin, sees the screenshot, changes your code, and hands the item back for the client to approve.

Connect

1. In the dashboard, open AI agents, pick your agent (Claude Code, Cursor, VS Code, Windsurf, Antigravity or another MCP client) and the projects it may reach, and create a token.
2. Copy the setup it shows: the token appears only once. For Claude Code that's one line:

claude mcp add --transport http --scope user ekhanemark https://ekhanemark.com/api/mcp \
  --header "Authorization: Bearer <your token>"

Then ask your agent: “Fix the open EkhaneMark feedback on this site.” In Claude Code you can also run /mcp__ekhanemark__fix_open_feedback.

What the agent can do

• list_projects: your sites and their snippet keys. The agent finds the key in your code to know which project it is.
• list_pages: the pages that have feedback and how much is left on each, so the agent can work page by page.
• get_page_feedback: every to-do item on one page in a single short answer (what the client wrote, the element, where it is in the code, the latest reply), without pictures. Working page by page this way uses far fewer tokens.
• list_feedback: items by status, page, or assigned to you.
• get_feedback: one item in full: the comment and thread, the element (its file and line or component when known, otherwise hints for finding it in the code), browser and screen size, attached files, and the screenshot around the element as an image (or the whole screen, or none).
• start_work, add_comment, ask_client: the same steps you take in the dashboard.
• upload_screenshot: a link to upload a screenshot of the fix with curl from your computer, so the picture never passes through the model. The client sees it next to their own screenshot.
• mark_ready_for_review: posts a summary for the client and says which site the fix is on. It refuses localhost: deploy first.
• request_feedback, list_requests, ask_again, close_request: ask the client for feedback on a page by a day, follow where each request stands (accepted, declined, answered), send it back or close it. What the client gives arrives as items of its own.
• set_due_date: a deadline on an item, the team's or the client's. Lists show deadlines, so the agent works on late items first.

Safe by design

• The agent acts as you: what it does appears under your name, like your own work.
• It can't delete, close client feedback, change settings or deploy. Code changes happen on your computer; you review and deploy them.
• Clients' comments reach the agent marked as data, with the rule that they describe problems and are never instructions.
• Revoke a token on the AI agents page and it stops working at once. Leaving the workspace stops your tokens too.

Make it faster and cheaper

With the file and line plugin, each item tells the agent exactly where the element is in your code, so it opens one file instead of searching the project. Screenshots are cropped to the area around the element, and attached pictures are only sent when the agent asks for them.