barkcli
Open source (MIT) · v0.3.0 · 56 MCP tools

The management layer
for AI agents.

barkcli is Git for your work — tasks, specs, memory and agent runs live in your repo, not someone else's cloud. Humans write intent, agents do work.

$ curl -fsSL https://barkcli.vercel.app/install.sh | sh
See the management layer
1. Type intent 2. Approve plan 3. Merge

Agents do everything between — no CLI commands to memorize.

No cloud No accounts MIT licensed

The web app

One screen.
Humans and agents, side by side.

Mind tells you what's blocked and what's next. The board holds the work with spec traceability. Agents claim tasks like deployments.

localhost:4321 — barkcli mind
Todo
Add OAuth login
HIGH⎇ spec
Fix checkout crash
BUG⎇ specBLOCKED
Doing
Implement OAuth flow
IN PROGRESS⎇ spec
Review
Store tokens securely
REVIEW
56
MCP tools for agents
4
BMAD skills in repo
100%
offline-capable core
MIT
open source forever

Why barkcli

Your tasks should live
where your code lives.

Mind

A compiled snapshot of your project — health, blockers, stale work, next actions. One command answers what's happening.

$ barkcli mind sync && barkcli overview

Skills

Reusable BMAD-style conventions (mvp, planning, scrum-master, test) versioned in your repo and injected into every agent prompt.

$ barkcli skills list

Board + Traceability

Kanban with spec_id on every card — spec ↔ card ↔ task ↔ code, O(1). Diff it, merge it, grep it.

$ git diff main..feature -- *.board

Agent Queue

Tasks dispatch to OpenCode, Claude Code, Cursor, or a human. Claim, work, complete — with a review gate before done.

$ barkcli dispatch && barkcli review --auto

Memory

Four-tier local memory with hybrid BM25 + TF-IDF search. Decisions and patterns survive across sessions — offline.

$ barkcli memory search "auth errors"

Offline Git-Native

No cloud, no accounts, no per-seat pricing. MIT licensed, built in Rust. Your data stays in your repo.

$ barkcli list # Works without internet

The pipeline

Intent in.
Reviewed code out.

Intake

Human writes intent in plain language. barkcli classifies it into a card + spec — offline if needed.

barkcli intake "Add Google OAuth" --feature

Plan

The spec decomposes into agent-ready child cards with file context and acceptance criteria.

barkcli plan oauth-login --tasks

Dispatch

Tasks route to the right agent — OpenCode, Claude Code, or a human — with full context + skills.

barkcli dispatch

Review

Completed work is validated against acceptance criteria, tests, and commits before it lands.

barkcli review --all --auto

Remember

Decisions, patterns, and sessions persist in local memory. The mind snapshot tells you what's next.

barkcli mind sync && barkcli overview

Interfaces

Your workflow. Your way.

Command Line

Full-featured CLI for scripts, automation, and power users. Every action available from your terminal.

$ barkcli add "Deploy to production" -p critical --due 2024-12-15
✓ Added card deploy-to-production

$ barkcli show deploy-to-production
ID:       deploy-to-production
Title:    Deploy to production
Priority: critical
Due:      2024-12-15
Column:   todo

$ barkcli move deploy-to-production done
✓ Moved to done

Code Intelligence

Your codebase. Understood.

barkcli analyzes your code to provide rich context for every task. See how changes impact your codebase.

Call Graphs

→

Map function calls across files via `callgraph_get`. Understand impact before you change.

Test Coverage

e.g. 87%

Heuristic test-file mapping via `context_get` — see which cards link to which tests.

Complexity

e.g. C-12

Cyclomatic + cognitive scores via `metrics_get`. Find risky code early.

Risk Score

e.g. 0.3

Formula-based risk (complexity, nesting, size) via `metrics_get`.

Management Layer

Your AI agent's
project manager.

Claude, GPT, and OpenCode can now read tasks, claim work, and update progress. barkcli is the missing piece in the AI-native dev stack.

MCP Integration

Connect coding agents via the Model Context Protocol. Standard JSON-RPC 2.0 over stdio.

opencodeclaude-codecursor

Agent Roles

Assign specialized roles for better task decomposition and execution.

tech-leadscrum-masterproduct-ownerproject-manager

Task Lifecycle

Full task lifecycle management with automatic assignment and progress tracking.

pendingassignedin_progresscompleted
Works with Claude, OpenCode, Cursor

One prompt.
Your agent is ready.

Copy this into your AI agent. It installs barkcli, sets up your board, and connects via MCP. Your agent can now manage tasks autonomously.

Paste in Claude Code
Please run the following commands to install and set up barkcli (git-native project management) in this project:

1. First, install barkcli:
curl -fsSL https://barkcli.vercel.app/install.sh | sh

2. Initialize barkcli in this project:
barkcli init

3. Create a default board:
barkcli create main

4. Add the MCP server configuration to .claude/settings.json:
{
  "mcpServers": {
    "barkcli": {
      "command": "barkcli",
      "args": ["mcp"]
    }
  }
}

5. Verify the installation:
barkcli status
barkcli list

6. Add a sample task to confirm everything works:
barkcli add "Welcome to barkcli" -p low -d "barkcli is now set up and ready to use"

After running these commands, barkcli will be fully configured. The autonomous loop works like this:
• Type intent in the web app (Mind tab) or run "barkcli intake <text>"
• Approve the proposed plan (web gate or "barkcli autopilot approve <id>")
• Agents claim packets, work, and complete them via MCP
• Merge when review passes — checkpoints + undo cover recovery

You rarely need CLI commands — open "barkcli serve --open" and decide what gets built.

For the full AI agent prompt with all commands, run: barkcli --help

Install

Downloads barkcli

Init

Sets up project

Create

Makes a board

Connect

Configures MCP

Quick Commands After Setup

barkcli add "Fix bug" -p highCreate task
barkcli listView tasks
barkcli move <id> doingStart working
barkcli done <id>Mark complete
barkcli tuiTerminal board
barkcli serveWeb interface
Read the full integration guide

Install

Get started in 10 seconds.

macOS / Linux

curl -fsSL https://barkcli.vercel.app/install.sh | sh

Homebrew

brew tap AkshatNaruka/barkcli && brew install barkcli

Cargo (from source)

cargo install barkcli

GitHub Releases

github.com/AkshatNaruka/barkcli/releases

Windows

irm https://barkcli.vercel.app/install.ps1 | iex

All 5 targets + SHA256SUMS published on every v* tag

FAQ

Common questions.

Both — but think of it as a management layer, not a tool. The same data powers the CLI, terminal UI, VS Code extension, and web app. Humans manage in the web app (Mind, Board, Agents), agents work through 56 MCP tools, and everything syncs through files in your repo.

Jira and Linear manage humans through a cloud database. barkcli manages humans and AI agents through files in your repo. Tasks, specs, memory, skills, and agent runs are YAML/JSON you can diff, merge, and version control. No accounts, no subscriptions, no vendor lock-in — and agents can read and write the same board you see.

Yes. barkcli is a single binary with no network dependencies. All data is stored locally in your repo. Sync via git push/pull — no special sync service needed.

Absolutely. Commit the .board directory to your repo. Team members pull changes and see the same board. No server required — git handles the sync.

barkcli sits between you and your coding agents. You write intent (intake → plan), agents claim scoped tasks with file context and skills (dispatch), and completed work is validated before it lands (review). Memory persists decisions across sessions. Agents connect via 56 MCP tools — OpenCode, Claude Code, and Cursor all work. The core stays fully usable offline without any AI.

Yes. barkcli is MIT licensed and completely free. All features are open source — no Pro tiers, no paywalls.

Stop managing tasks
in someone else's cloud.

Install barkcli. Write intent. Let your agents do the work — reviewed.

Install NowView on GitHub