You captured a bug on a screen, dropped a pin, typed the fix you want, and hit Publish. Now your agent needs it. Cobalt Capture gives you three ways to hand the same review to a coding agent, and the wrong one wastes a turn. Here is which to use.
What's the difference between the /r/ link and the /markdown url?
The /r/<slug> link is the full review: screenshots, pins, your comments, and a reader that people open in a browser. The /r/<slug>/markdown URL is the same review as plain text, structured so a machine can parse it.
Here is the part that trips people up. You usually do not need to pick. When an AI assistant opens the /r/<slug> link, Cobalt serves it the markdown automatically. So if you paste the review link into Claude or ChatGPT, the model gets the clean version without you touching the /markdown URL. The explicit /markdown URL is there for when you want to be certain, or when you are piping the text into a script. For the background on what that output contains, see how markdown screenshots work.
So when do I actually paste a url versus connect over mcp?
Paste a URL when you have one review and want it read once. Drop the /r/ link into the chat, say what you want done, and the agent reads the review including the screenshots.
Connect over MCP when you want the agent to pull reviews itself, read screenshots inline, and write new ones back, with no copy and paste between tabs. Cobalt's MCP server is the standard way agents connect to tools, and once it is added, your agent can fetch a review by slug on its own. The setup takes one line. Follow the steps to connect your coding agent over MCP and you will not paste a URL again for routine work.
Short version: a URL is a one-off, MCP is a standing connection.
Which one works in cursor?
Both. You can paste a /r/ link into Cursor's chat and it will read the review. Or you can add the MCP connector and pull a review into Cursor over MCP without leaving the editor. MCP is the cleaner path if you review often, because Cursor fetches the screenshots directly instead of chasing a link. The same choice applies in Claude Code, Codex, and any assistant that supports MCP connectors.
Will the agent actually see my pins and arrows?
Yes. Pins, arrows, boxes, and labels are baked into the exported images, so whatever you marked is visible in the screenshot the agent reads. This matters more than the URL you choose. A pin on the exact button beats a paragraph describing where the button is. That is why a review built in Cobalt tends to land in one turn: the agent gets the picture and the marked-up spot together, not a vague description. If you want the agent to edit the right element, crop tight and pin it, as covered in cropping a still for an agent.
Can the agent read a review someone else published?
If they sent you the /r/<slug> link, yes. Anyone with the link can read the review, and an assistant opening that link gets the markdown. You do not need to own the review to feed it to your agent. You do need to own it to have the agent write back through MCP or to send it to a GitHub repository as an issue.
Does any of this cost money?
Trying it is free. Unlimited reviews and screenshots, no password and no credit card; Cobalt just asks for an email when a review is ready so it can send you the link. One catch worth knowing up front: after 7 days a review goes read-only for you, which stops editing, exports, and agent access. The shared link keeps working for the people you sent it to. Pro is $5 a month or $50 a year, keeps every review editable and reopens read-only ones instantly, and cancels any time. The full breakdown is on the pricing page.
I just want the fastest path. What do I do?
If this is a one-time thing: publish the review, copy the /r/ link, paste it into your agent's chat, tell it the fix. Done. If you hand reviews to an agent more than once a week: spend the one line to add the MCP connector and skip the pasting from then on. Either way, start a new review, mark the exact spot, and let the screenshot carry the detail your agent needs.