Your coding agent asks Lissoss which page needs work, reads the evidence, pulls the brief — and after the deploy, sees whether the numbers moved. Claude Code, Cursor, VS Code, Windsurf and Claude Desktop.
Which page on our site needs work first?
- lissoss_pages_needing_attention(window: 30d)
Next: the implementation brief for /pricing
MCP — the Model Context Protocol — is how AI coding tools plug into other services. Lissoss's MCP server is that plug for your visitors: the agent already working in your website's code can ask Lissoss what they do, and what to change.
- 01Lissoss finds itWhich page loses visitors, where they rage-click, and one fix for each — from your own traffic.
- 02Your agent asksIn plain words, from your editor. It reads the evidence and pulls the brief — nothing to copy across.
- 03It ships, Lissoss checksThe agent makes the change. After the deploy, Lissoss says whether the numbers moved.
One address, one key, four steps — and the app checks the third for you.
- 01
Make the workspace's key.
In Lissoss, an admin opens Settings → Coding agents and creates the key. It is shown once — and that page puts it straight into the snippet for your tool, so there is nothing to retype.
- 02
Add Lissoss to your tool.
Pick the tool you code with. Every one takes the same two things: Lissoss's address and your key.
Terminal, in your website's repository claude mcp add --transport http lissoss https://your-lissoss-address/api/mcp \ --header "Authorization: Bearer lk_your_key"That saves Lissoss for this repository, on your machine only. To have it in every project, add --scope user.
- 03
Check it's connected.
Restart your tool and it lists lissoss with its eight tools. In Lissoss, Settings → Coding agents turns green the moment your agent connects, and names the tool it saw.
Claude Code, in a terminal claude mcp list # lissoss … ✔ Connected - 04
Ask in plain words.
Your agent picks the tools itself. Seven are free; lissoss_run_review uses one AI action and says so, so your agent can ask you first.
First prompts, in your website's repository Ask lissoss which page needs the most work, and why. Pull lissoss's implementation brief for that page and make the high-priority changes. Ask lissoss whether the fixes we shipped worked.
What a coding agent can ask Lissoss, once the workspace's key is in its MCP settings.
| Tool | What it does |
|---|---|
| lissoss_overview | How is the site doing? Stats, top landing and exit pages, frustration, sources. |
| lissoss_pages_needing_attention | Where should I focus? Attention scores with their reasons. |
| lissoss_page_profile | What happens on one page: clicks, rage and dead clicks, flows, devices. |
| lissoss_implementation_brief | What exactly to change: the stored review as a task list. |
| lissoss_wireframe | The target page structure as JSON, for code or Figma. |
| lissoss_issues | The Issue Inbox, with fixes and verification verdicts. |
| lissoss_experiments | Stored A/B test proposals. |
| lissoss_run_review | A fresh page review, stored for the dashboard too. Uses one AI action. |
Sample data: what a coding agent asks the Acme workspace, and what it hears back.
- Asks
- Which page needs the most work?
- lissoss_pages_needing_attention
- /pricing — attention 86 · frustration in 28 sessions
- lissoss_implementation_brief
- Make the Enterprise card a real link to /contact (or add a “Talk to us” button inside it) so it does what its styling promises.
- lissoss_issues
- “Compare plans” did nothing on mobile — improved, −85%
On every plan, and in the 14-day trial.
- Cost
- Seven tools are free. lissoss_run_review uses one AI action — and says so, so your agent can ask you first.
- Access
- It reads; it doesn't write. Marking an issue fixed stays with you, in the app.
- Keys
- One per workspace, made by an admin. A wrong key is refused, every time.
- Rate limits
- Up to 120 requests a minute per key, and four reviews a minute.
No card. One snippet. Sample data until yours arrives.