Cclarity
A LinkedIn MCP for Claude and ChatGPT that I built and ran from 2023 to 2026. It ranked engagers by ICP fit inside the AI you already use. Now being rebuilt as a web app.
What it is
Cclarity was a LinkedIn MCP for Claude and ChatGPT that I built and ran from 2023 to 2026. It put your real LinkedIn data, the names, roles and ICP fit behind every like, comment and profile view, inside the AI you already used.
You installed it once. Your AI then saw who engaged on your posts, who viewed your profile, and which of them matched your ideal customer profile. The promise was to replace a $2K LinkedIn agency with the AI on your desk. The original stack was retired in September 2026 and I am rebuilding the product as a web app.
Why an MCP, not another dashboard
Most LinkedIn tools dump a dashboard on you and leave you to interpret it. Cclarity did the opposite. It exposed your LinkedIn data as a Model Context Protocol server, so Claude or ChatGPT can read it directly in conversation.
Ask “who engaged on my last post and matches a Series A founder ICP?” and your AI answers with names, roles, and a ranked list. Ask it to draft a follow-up DM, and it uses the engager’s actual profile to write it. The intelligence lives where you already think and write.
Read-only by design
Cclarity never posted, commented, or sent DMs on your behalf. No scraping, no automation, no flagged accounts. It only read data you can already see in LinkedIn, through LinkedIn’s authorised flow. Read-only was why your account stayed safe, and it was why we told users to send the message themselves. The intelligence is the product; the sending is yours.
How it was built
Built with a small team. I handled product direction and contributed frontend code alongside an engineer on the backend. The MCP server ran on api.cclarity.io/mcp. It worked in regular Claude (web or desktop), regular ChatGPT (Plus, Pro, Team), Claude Code, and Codex.
What I learned
Distribution beats features. A dashboard with twenty charts loses to a single tool call that drops the right five names into the conversation you are already having. Build for where your user already is.
See my other projects: CheckHowMuch.sg (9,700+ pages of property data, built with Claude Code) and SeeWhatIf (134 healthcare cost scenarios, also built with Claude Code).