Persnally is an MCP server that learns from your conversations with Claude. It builds a private interest graph and sends you a personalized digest — no setup, no surveys, just chat.
you: I'm building a real-time data pipeline with Kafka and Rust. Trying to decide between rdkafka and kafka-rust...
claude: Let me compare both libraries for your use case...
persnally tracked: Rust async programming (0.9), Kafka data pipelines (0.8), systems architecture (0.5)
No raw messages stored. Only structured signals.
Persnally runs silently alongside your AI conversations. No behavior change required.
One npm install, add to your Claude Desktop config. Takes under 2 minutes. Persnally runs locally on your machine.
As you discuss topics with Claude, Persnally extracts structured signals — topics, intent, sentiment, depth. Claude IS the NLP engine.
Daily or weekly, Persnally curates content matched to your interest graph and sends it via email. Real links, real articles, zero filler.
Every other recommendation engine asks you to fill out surveys, rate things, or connect accounts. Persnally doesn't need any of that.
When you talk to Claude about Rust async programming, startup fundraising, or LLM fine-tuning — Claude already understands the context, sentiment, and depth. Persnally just gives it a structured way to report what it observed. Zero extra AI cost. Zero NLP pipeline.
Your conversations are the most honest signal of what you care about. Not what you say you like — what you actually spend time on.
Not a static preference list. A weighted, decaying, sentiment-aware graph that evolves as you do.
Topics you discussed months ago fade naturally. What you talked about yesterday carries more weight. 7-day half-life keeps the graph fresh.
"I hate CSS" and "I love CSS" are very different signals. Negative sentiment deprioritizes topics — no frustrated content in your digest.
A brief mention of React scores differently than a 30-minute deep dive on React Server Components. Depth matters.
Your digest covers multiple interest categories proportionally. No echo chambers — technology, business, career, science all get fair weight.
"React.js", "React JS", and "ReactJS" all map to the same node. The graph handles synonyms so you get clean, deduplicated interests.
Are you learning, building, researching, or debugging? Intent shapes what content gets surfaced — tutorials vs deep dives vs release notes.
Not privacy by policy. By architecture. Raw conversations never leave your machine.
Your interest graph is a JSON file on your machine. You can read it, edit it, or delete it anytime.
Use persnally_forget to remove any topic, or clear everything.
Install the MCP server and add it to Claude Desktop.
{
"mcpServers": {
"persnally": {
"command": "persnally",
"args": []
}
}
}That's it. Persnally will start learning from your conversations automatically. Ask Claude "show my Persnally interests" anytime to see what's been tracked.
Persnally exposes 5 MCP tools that Claude calls automatically during your conversations.
persnally_trackExtracts topics, intent, and sentiment from the current conversation. Called automatically by Claude.
persnally_interestsShows your current interest profile — what Persnally has learned, grouped by category.
persnally_digestGenerates and sends your personalized digest email based on your interest graph.
persnally_configSet your email, digest frequency (daily/weekly), and API preferences.
persnally_forgetRemove a specific topic or clear all data. Your data, your control.
Persnally is fully open source under MIT. Your interest graph lives on your machine. No vendor lock-in, no data silos.