devrecall
DevRecall is a local-first, privacy-focused MCP server that gives AI coding tools (Claude Code, Cursor, Codex, etc.) memory of your developer activity, stored in a local SQLite database. It aggregates activity from Git, GitHub/GitLab/Bitbucket, Slack, Google Calendar, Jira/Linear, and Confluence.
Track & Log Activity
Automatically collects commits, PRs/MRs, reviews, issues, Slack messages, meetings, ticket transitions, Confluence pages, and more
Manually log decisions, conversations, and notes with people/topic tags
Query & Search
List activities by time window, source, type, or person
Count and aggregate activities (group by source, type, identity, day, or week)
Keyword search (FTS5 full-text) or semantic/vector similarity search for vague queries
Fetch full content and metadata for a specific activity by ID
Understand Work Context
Find all activities sharing a ticket key (commits, PRs, pages, etc.)
Look up work by a specific person with optional keyword and date filters
Surface recent decisions (ADRs, RFCs, decision-tagged notes)
Meeting Prep
Get event details and each attendee's recent activity for an upcoming meeting
Reports & Summaries
Access pre-built daily, weekly, monthly, or quarterly summaries
Utilities
Resolve person identities (Git authors, Slack users, etc.)
Get current UTC time to anchor relative date references like "last week"
Collects pull requests, reviews, issues, and comments from Bitbucket.
Collects pages, blogposts, and comments you authored from Confluence.
Collects commits, branch activity, and changed files from local Git repositories.
Collects pull requests, reviews, issues, and comments from GitHub.
Collects merge requests, reviews, issues, and comments from GitLab.
Collects meetings attended, organized, and declined from Google Calendar.
Collects issue transitions, comments, and sprint membership from Jira.
Collects issue transitions, comments, and sprint membership from Linear.
Collects messages and threads you participated in from Slack.
DevRecall
Your developer activity, aggregated on-device. No cloud sync. No telemetry. Your data never leaves your machine.
DevRecall pulls from Git, Slack, Google Calendar, Jira, Linear, Confluence, GitHub/GitLab/Bitbucket, Notion, and your Claude Code sessions; stores it in a local SQLite database; and turns it into standups, weekly reports, brag docs, and a chat that actually knows what you worked on.
Also ships an MCP server —
any MCP-compatible coding tool (Claude Code, Cursor, Codex, Continue, Zed)
can spawn it as a stdio subprocess and gain memory of everything you've
shipped. /devrecall:recall what auth bug did I fix in February returns
cited commits, PRs, and tickets inline.
📚 docs.devrecall.dev — install, configure, integrations, CLI reference.
Why
Local-first. SQLite on your laptop. Tokens in
~/.devrecall/tokens/(0600). The Cloudflare Worker relay is a pass-through for OAuth callbacks only — it never sees your data.LLM-optional. Bundled embeddings run offline (ONNX +
all-MiniLM-L6-v2). Use local Ollama for chat, or bring your own OpenAI/Anthropic key.Open source. MIT-licensed. Audit it, fork it, build it from source.
Related MCP server: GoodMemory
Sources
Source | What gets collected |
Git (local) | Commits, branch activity, files changed |
GitHub / GitLab / Bitbucket | PRs/MRs, reviews, issues, comments |
Slack | Your messages, threads you participated in |
Google Calendar | Meetings attended, organized, declined |
Jira / Linear | Issue transitions, comments, sprint membership |
Confluence | Pages, blogposts, and comments you authored |
Notion | Pages you created or edited |
Claude Code | Coding-agent sessions, linked to repo & branch |
Install
macOS — desktop app + CLI:
brew install --cask pavelpilyak/devrecall/devrecallInstalls DevRecall.app and puts the devrecall CLI on your PATH.
Linux / WSL — CLI and MCP server (no desktop app):
brew install pavelpilyak/devrecall/devrecall-cliWithout Homebrew, grab the tarball directly:
curl -L https://github.com/pavelpilyak/devrecall/releases/latest/download/devrecall-linux-x86_64.tar.gz \
| tar xz && sudo mv devrecall-linux-x86_64 /usr/local/bin/devrecallSupported on Ubuntu 22.04+ and equivalent (Debian 12+, Fedora 35+).
Windows — no native build. Everything except the desktop app runs under WSL using the Linux instructions above, including the MCP server, which your editor can spawn across the WSL boundary.
git clone https://github.com/pavelpilyak/devrecall.git
cd devrecall
make build # → bin/devrecallRequires Go 1.22+ with CGO enabled (for SQLite FTS5).
Then connect a source and generate your first standup:
devrecall auth github # or: slack, google, jira, linear, notion, gitlab, bitbucket
devrecall standupFull install + setup walkthrough at docs.devrecall.dev/install.
Development
make build # bin/devrecall
make test # tests with race detector
make lint # golangci-lintBuild tags: fts5 enables SQLite full-text search; GO enables hugot's pure
Go ONNX backend for embeddings.
The desktop app (Tauri + Svelte) lives in desktop/;
the OAuth callback relay (Cloudflare Worker) in relay/.
Contributing
Bug reports and collector contributions are especially welcome. See CONTRIBUTING.md.
Security
SECURITY.md for responsible disclosure.
Available Tools
17 toolscount_activitiesA
Count activities in a time window, optionally grouped by source, type, identity, day, or week. The primitive behind 'how many', 'how often', and 'compare X vs Y' questions.
Inputs: end, type, start, source, +2 more.
current_timeA
Return the current time, in the user's local timezone and in UTC. Prefer today for date filters — it is the user's local calendar date, which is what 'today' means to them.
Inputs: no named parameters.
get_activityA
Fetch the full activity row (including content and metadata) for a given activity ID. Use after list_/search_ to drill into a specific row.
Inputs: id (required).
get_summaryA
Fetch a single pre-built summary by period type and start date. Returns null if no summary exists for that period.
Inputs: period_type (required), period_start (required).
get_work_itemA
Fetch a work item (a ticket or PR and everything linked to it) with its full cross-source timeline: the ticket, its commits, PRs, reviews, and discussions in chronological order. The best tool for 'show me everything about PROJ-123'.
Inputs: id, key, limit.
list_activitiesA
List activities in a time window with optional source/type/person filters. Returns shallow rows (no body); use get_activity to fetch full content.
Inputs: end, tag, type, limit, +4 more.
list_identitiesA
List people known to the system (Git authors, Slack users, etc.), optionally filtered by a name/email substring.
Inputs: query.
list_summariesA
List pre-built periodic summaries (daily/weekly/monthly/quarterly). Prefer these over re-summarizing raw activities when answering 'summarize my Q1' or 'what happened last week'.
Inputs: period_type (required), end, limit, start.
list_work_itemsA
List work items (tickets/PRs with their linked activity), most recently active first. The structural answer to 'what was I working on last week' — each item groups a ticket with its commits, PRs, and discussions. Use get_work_item for the full timeline of one item.
Inputs: end, limit, query, start, +1 more.
log_eventA
Record a manual note — a decision, in-person conversation, observation, or anything else worth remembering. Stored as type=note from source=manual and indexed alongside everything else. Use this when the user dictates context you want recallable later.
Inputs: text (required), tags, people.
prep_meetingA
Brief for a calendar meeting: returns the event details plus each attendee's recent shipped activity. Look up by activity_id if known, otherwise by date+title_contains. Use this to walk into a 1:1 or sync knowing what everyone's been doing.
Inputs: date, activity_id, lookback_days, title_contains, +1 more.
recent_decisionsA
Find decision-shaped activity in a date range: manual notes (type=note) merged with anything whose title or body hits decided/decision/RFC/ADR. Use this before answering 'what did we decide about X' or composing a quarterly summary.
Inputs: end, limit, start, source.
resolve_personA
Resolve a name or email to a single identity (the closest match). Returns the matched identity or null if no candidate is found.
Inputs: name_or_email (required).
search_activitiesA
Hybrid search over activity titles and content, combining keyword (FTS5) and semantic similarity. This is the default search tool — it handles both concrete phrases like 'retry strategy' and loosely-worded queries, so prefer it unless you specifically need pure vector similarity.
Inputs: query (required), end, tag, limit, +2 more.
semantic_search_activitiesA
Pure vector similarity search over activities. search_activities already blends this with keyword matching, so reach for this only when that returns nothing useful and the query is purely conceptual with no reliable keywords. Returns shallow rows ranked by similarity.
Inputs: query (required), end, limit, start.
who_worked_onA
Activities authored by a specific person, optionally narrowed by a keyword query and date range. Pass identity_id if you have one (from list_identities/resolve_person); otherwise pass name_or_email and the tool resolves it.
Inputs: end, limit, query, start, +3 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 17 tools
Most tools have clearly distinct purposes—search, list, count, drill-down, and work-item aggregation are separated well. The main overlaps are get_related_activities vs get_work_item, and search_activities vs semantic_search_activities, but the descriptions provide enough guidance to choose correctly.
The core set follows a consistent verb_noun snake_case pattern like list_activities, count_activities, and resolve_person. A few names deviate—who_worked_on, recent_decisions, current_time, and the abbreviated prep_meeting—but the overall style remains readable and predictable.
At 17 tools, this is slightly above the ideal 3-15 range, but the server covers a broad domain spanning activity search, work items, people, summaries, and meeting prep. Most tools earn their place and support distinct workflows rather than padding the surface.
The read/query surface is strong: search, list, count, drill-down, person resolution, work-item aggregation, and pre-built summaries cover the main recall workflows. The notable gap is that log_event supports creating manual notes but there is no update or delete path, and there is no explicit metadata tool for enumerating sources or types.
Maintenance
Related MCP Connectors
Team memory for AI coding agents and code reviews, reviewed by humans and served over MCP.
Your own memory for every AI: Claude, ChatGPT, Cursor, Zed and more. Brings in 20+ sources.
Shared memory for coding agents. Stop re-explaining your codebase every session.
A second brain for your AI agent. Shared memory for your team's AI agents, plus public and private knowledge bases — works with ChatGPT, Claude, Cursor and any MCP client.
Related MCP Servers
- AlicenseAqualityBmaintenanceSemantic codebase search + persistent working memory for AI code editors. Local, zero-config, MCP. No API key.2344 PyPI2MIT
- AlicenseBqualityBmaintenanceLocal-first, auditable memory for Codex, Claude Code, and MCP clients. It stores scoped user/project memory in SQLite or Postgres, serves read-only recall and inspection tools by default, and supports opt-in governed writeback with review and forget controls.8159 npm21MIT
- FlicenseNot gradedqualityCmaintenanceLocal-first memory for AI coding agents. Indexes your Claude Code, Codex, Cursor, Cline and Gemini CLI history by repo/branch/worktree and hands a context capsule (decisions, touched files, next steps) to the next agent over MCP.-
- FlicenseNot gradedqualityAmaintenanceEnables AI coding agents to store, search, and retrieve persistent local-first memory — project-specific decisions, preferences, error patterns, and research plus cross-project guidelines — so context survives across sessions without any external API calls. Optionally also exposes a task queue for autonomous execution and GitHub issue resolution in MCP clients such as Claude Code, OpenCode, and Goose.-