LabscoMCP early access ↗
pavelpilyakMCP SERVER

DevRecall

Aggregate your Git, Slack, calendar, tickets and coding sessions into a local SQLite database, then ask what you worked on and get cited commits, PRs and tickets back.

by pavelpilyak ↗Session & Transcript RecallCommunity

LABSCO SUMMARY

Local-first is the whole proposition, and the architecture backs it.

Plenty of tools promise to remember your work; this one does it without your Slack messages and calendar leaving the laptop, down to embeddings that run offline and a relay that only forwards OAuth callbacks. That matters because the data being aggregated here — every message, meeting and ticket you touched — is exactly the data you would not want pooled somewhere else. The trade is that setup is per-source and the index is yours to maintain.

What it is

A developer activity aggregator that runs entirely on your machine, with an MCP server on top. It collects from your tools into a local SQLite database and turns that into standups, weekly reports, brag docs and a chat that can answer questions about your own work with citations. No cloud sync and no telemetry.

What you get

  • Commits, branch activity and changed files from local Git; PRs, reviews, issues and comments from GitHub, GitLab or Bitbucket
  • Your Slack messages and the threads you took part in; meetings attended, organised and declined from Google Calendar
  • Issue transitions, comments and sprint membership from Jira or Linear; pages, blogposts and comments you authored in Confluence
  • Your coding-agent sessions, linked back to the repository and branch they happened on
  • Recall from any MCP-compatible coding tool: ask what you fixed in a given month and get the commits, PRs and tickets cited inline
  • Written output on demand — standups, weekly reports and brag docs assembled from what actually happened
  • Embeddings that run offline (ONNX with a local sentence-transformer model), with chat through local Ollama or your own OpenAI or Anthropic key

Requirements

Accounts you already have, connected one at a time — devrecall auth github, and the same for slack, google, jira, linear, gitlab or bitbucket. Tokens are stored under your home directory with restrictive permissions, and the hosted relay exists only to pass OAuth callbacks through, never your data. Install on macOS with Homebrew, which puts the CLI on your PATH alongside the desktop app; building from source needs Go 1.22+ with CGO enabled for SQLite full-text search. The MCP server is spawned as a stdio subprocess by any compatible tool. MIT licensed.

Setup effort

Follow the published steps — brew install --cask pavelpilyak/devrecall/devrecall

See installation options