M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
B

Backlog.md

Community
Markdown-native Task Manager & Kanban visualizer for any Git repository
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #12 of 438
Stars
★ 7.0k Very popular
Transport
stdio (local process)
Runtime
Node.js · Bun · Prebuilt binary
Credentials
No credential needed
License
MIT
Last commit
Tools
2
72FMRS · B

Backlog.md is a Markdown-centric task manager designed for AI-agent collaboration. It provides a structured workflow including task decomposition, plan approval, and code review. It supports terminal and web UIs. MCP integration is available for multiple clients and supports workspace following via MCP roots. Potential risks include configuration complexity and dependence on Git operations, but overall it offers a lightweight, local-first alternative to centralized task managers.

Strongest · Documentation 17/20 Weakest · Reliability 11/20

Reliability
11/20
Security and permissions
14/20
Maintenance
17/20
Documentation
17/20
Setup experience
13/20
Why each score
Reliability 11/20
The project is a mature CLI tool with a well-documented MCP server startup command (`backlog mcp start`) and client integration commands. However, there is no separate server manifest file, and no dedicated MCP server tests or explicit CI evidence verifying the MCP handshake. The primary reliability evidence comes from the CLI's maturity and widespread use (dogfooded), but static review cannot confirm actual MCP initialization and tool-list behavior. Therefore, per static calibration, award no more than 12, and deduct for lack of tests and CI evidence specific to the MCP path.
Security and permissions 14/20
The tool is local-first by default, with no telemetry, and data is stored in local files only. The MCP server is started via `backlog mcp start` and supports `BACKLOG_CWD` and `--cwd` to pin the project. Documentation emphasizes that remote Git operations are optional and the server only operates within the client's MCP roots. No sensitive credential handling or external network calls were found. However, permission and confirmation mechanisms are not explicitly documented for MCP, and dangerous operations like rename/delete may require confirmation but are not clearly mentioned. Thus, no red lines triggered, but explicit least-privilege and confirmation statements are lacking, giving 14.
Maintenance 17/20
The repository is highly active, with 6414 stars, 37 open issues, and not archived. README shows frequent version updates (e.g., v1.40) and mentions progress. MIT license is explicit. Evidence of sustained maintenance includes release history, community tools (VS Code extension), and public talks. No security-response channel seen, but overall strong maintenance, giving 17.
Documentation 17/20
Documentation is comprehensive, including full CLI reference, advanced configuration guide, troubleshooting (e.g., Apple Silicon) and detailed MCP client configuration examples (Claude Code, Codex, Gemini CLI, etc.). Workflow guides are also included. However, details on MCP tool parameters, limits, and costs are missing, and installation README lacks step-by-step for all platforms. Layered docs with sufficient examples are present, but some hidden assumptions (e.g., need to initialize project first) exist, thus deducting to 17.
Setup experience 13/20
Installation is simple: `npm i -g backlog.md` or `npx backlog.md`. Multiple platforms (macOS, Linux, Windows) are supported with Homebrew and Nix. MCP setup requires first running `backlog init` then configuring the client, with clear commands for each client. However, manual config requires environment variables or args, and troubleshooting steps are not provided. Static review lacks execution evidence, so cap at 15, and complexity of external runtime (Node/Bun) increases fragility, giving 13.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local files

Best for

  • Developers who want to collaborate with AI coding assistants while keeping human oversight at every stage
  • Teams needing a lightweight, file-based, local-first task management solution
  • Projects aiming to improve AI output quality through clear acceptance criteria and definitions of done

Not for

  • Teams needing centralized, server-based task management with user roles and permissions
  • Users needing complex workflow automation or deep integration with proprietary tools
  • Projects requiring real-time multi-user synchronization

Required permissions

  • Read and write Markdown task files in the project repository
  • Read and modify local filesystem (tasks, configuration)
  • Optionally: perform Git operations (auto-commit, remote operations)
  • Run within the specified MCP roots or BACKLOG_CWD

Risks and side effects

  • Potential inconsistency in field types or metadata if task files are hand-edited
  • Git operations may be required for some features; non-Git projects are limited to local operations
  • If the project is not initialized, the server may serve init-required resources until initialized
  • Cross-branch accuracy may be limited depending on configuration

Setup

Before you start

Runtime：Node.js · Bun · Prebuilt binary

Other optional settings (1)
BACKLOG_CWD optional Optional. Absolute path that pins the MCP server to one project and disables workspace following; alternatively pass --cwd as an arg.

Install globally: npm i -g backlog.md
Initialize in any Git repo: backlog init "My Project"
Or initialize with --no-git for non-code projects
Choose the MCP connector option to auto-configure Claude Code, Codex, Gemini CLI, Kiro, or Cursor
Alternatively, manually add the MCP server configuration (see install_config)
Optional: Add instructions in CLAUDE.md or AGENTS.md for agents to read backlog://workflow/overview

.mcp.json 
{
  "mcpServers": {
    "backlog": {
      "command": "backlog",
      "args": [
        "mcp",
        "start"
      ],
      "env": {
        "BACKLOG_CWD": "/absolute/path/to/your/project"
      }
    }
  }
}

Shown for Claude Code. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "backlog": {
      "command": "backlog",
      "args": [
        "mcp",
        "start"
      ],
      "env": {
        "BACKLOG_CWD": "/absolute/path/to/your/project"
      }
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add backlog -e BACKLOG_CWD=/absolute/path/to/your/project -- backlog mcp start

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

Run the /mcp command in your AI client (Claude Code, Codex, Kiro) to confirm the backlog server is connected. If no initialized project is found, the server serves backlog://init-required, so run backlog init first.

Troubleshooting

  1. If you receive 'illegal hardware instruction' or binary package not installed, ensure native architecture on Apple Silicon and reinstall (e.g., arch -arm64 npm i -g backlog.md)
  2. Verify MCP server connection: use /mcp command in your AI tool
  3. Run backlog config to check configuration and ensure project is initialized
  4. If MCP server cannot find the project, set BACKLOG_CWD or use --cwd argument
  5. Refer to CLI reference and advanced config docs (CLI-INSTRUCTIONS.md, ADVANCED-CONFIG.md)

Things to try

Once connected, you can ask your AI assistant things like:

  • Decompose my search-feature idea into small Backlog.md tasks with acceptance criteria
  • Show me the details and acceptance criteria for task BACK-10
  • Mark task BACK-3 as Done and add a comment about the UI split
  • List all tasks currently in the To Do status

Tools 2

mcp read-only
Launches the MCP server, providing Backlog.md workflow instructions and task management capabilities.
start read-only
Starts the MCP server, providing Backlog.md workflow instructions and task management capabilities.

Use cases

Manage AI-agent-driven software development projects with a spec-driven workflow
Decompose large features into small, reviewable tasks, each with its own PR
Review task descriptions, acceptance criteria, and implementation plans before coding
Track project milestones and dependencies
Visualize task progress in the terminal or browser (Kanban board)
Manage personal projects without Git

Supported clients

Claude Code
Codex
Gemini CLI
Kiro
Cursor

Listed from the project's documentation, not tested by this site.

Overview

Backlog.md turns any folder into a self-contained project board powered by plain Markdown files and a zero-config CLI. It is designed for AI-agent collaboration, offering structure for task descriptions, acceptance criteria, milestones, implementation plans, and review checkpoints. It provides a terminal Kanban, a web UI, fuzzy search, local-first storage, and MCP integration for tools like Claude Code, Codex, Gemini CLI, and Kiro.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 62.9k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 7e2450577fc0 Data synced 2026-10-12 Read the FMRS scoring method