memgrep is a local control plane for Cursor. Ingest chats from Cursor, Claude Code, and Kiro. Hybrid recall finds past fixes and playbooks on your machine. Coding loops run until exits pass, then can open a PR. Drive Cursor from Telegram. Cron your playbooks. One MCP server for memory, jobs, and loops, plus optional tools: Jira, Neon, gcloud, PostHog, Upstash, Product Hunt, and more. Open source. Lock workflows you already figured out. npm i -g memgrep · memgrep.getuigen.dev

Hey hunters, I used to live in Kiro, then moved to Cursor. My old sessions didn’t come with me, and I kept losing context I had already paid for in tokens and time. That pushed the first idea: a shared memory layer for agents that stays on my machine. Telegram came next. I wanted to kick off real Cursor work from my phone, with that memory attached mid-task. Once that worked, I stopped treating memgrep as “just recall” and started exporting more of my workflow into it: playbooks, jobs, MCP tools (Jira, gcloud, and friends). Loops showed up for a simple reason. I already review work against a PR checklist. PRs fail when they miss criteria. So I asked: why not make the agent loop on those rules until it passes? That became the coding loop: task in, exit conditions as review rules, then exit actions like opening a PR. memgrep is that stack in one place: local memory, loops that finish work, Telegram → Cursor, scheduled playbooks, and one MCP server for the tools I actually use. Open source. Happy to answer questions.

finally gave this a spin and the hybrid recall pulling up my old jira fixes when i hit a similar error was genuinely useful, saved me re-deriving a redis config i had already figured out last month.

memgrep

Local Cursor: memory, loops, MCP integrations

Local Cursor: memory, loops, MCP integrations