Supported AI Tools

Backlog.md supports integration with a variety of mainstream AI coding assistants. Connection methods vary slightly by tool, but the core principle is the same: register Backlog.md's MCP server with the AI client so it can directly invoke Backlog.md tools.

Supported Tools Overview

Claude Code

Claude Code is Anthropic's command-line AI assistant with native MCP support.

Run the following command to register:

claude mcp add backlog --scope user -- backlog mcp start

Parameter description:

• --scope user: writes the configuration at user level, available to all projects.
• backlog mcp start after --: the command to start the Backlog.md MCP server.

After successful registration, type @backlog in a Claude Code session to invoke Backlog.md tools.

OpenAI Codex

OpenAI Codex CLI also supports the MCP protocol.

Run the following command:

codex mcp add backlog -- backlog mcp start

Note that Codex uses -- as the stdio command separator. This format has been updated to match current Codex CLI behavior.

Codex automatically includes the Backlog.md toolset in its available tools list. During conversations, Codex autonomously decides when to call Backlog.md tools based on context.

Google Gemini CLI

Gemini CLI provides MCP support through the gemini command.

Run the following command:

gemini mcp add backlog -s user backlog mcp start

Parameter description:

• -s user: equivalent to --scope user, applies at user scope.
• backlog mcp start: startup command.

Kiro

Kiro CLI's configuration is slightly different and requires explicitly specifying how to split arguments.

Run the following command:

kiro-cli mcp add --scope global --name backlog --command backlog --args mcp,start

Note that arguments after --args are comma-separated, not space-separated.

Cursor

Cursor supports two ways to integrate with Backlog.md.

Method 1: MCP Configuration

Open Cursor settings, find the MCP configuration entry, and manually add the mcpServers configuration:

{
  "mcpServers": {
    "backlog": {
      "command": "backlog",
      "args": ["mcp", "start"]
    }
  }
}

After saving, Cursor's AI assistant can use the full set of Backlog.md tools.

Method 2: CLI Instruction File (AGENTS.md)

Cursor also reads the project root AGENTS.md as a general agent instruction. When running backlog init or backlog agents --update-instructions, the Cursor-specific instruction is written into AGENTS.md; a separate .cursorrules file is no longer generated. On repeated runs, the existing AGENTS.md content is preserved, and Backlog.md only updates its marked block.

GitHub Copilot

GitHub Copilot currently does not support the MCP protocol, so it is integrated via CLI instructions.

Run the following command to generate the agent instruction file:

backlog agents --update-instructions

This command generates .github/copilot-instructions.md in the project root, containing:

• Complete Backlog.md CLI command reference
• Task creation conventions
• Acceptance criteria format
• Workflow guide

Copilot reads this file when generating code to understand the project's task-management conventions and assist development better.

Verifying Integration

After configuration, verify integration as follows:

1. Start the corresponding AI client.
2. Enter a prompt such as "List my pending tasks".
3. Observe whether the AI correctly calls Backlog.md's list_tasks tool.
4. If the AI returns the current project's task list, integration is successful.

If integration fails, check:

• Whether the backlog command is in the system PATH.
• Whether backlog mcp start can start normally (test separately in a terminal).
• Whether the AI client's MCP configuration correctly points to backlog mcp start.
• If using Codex, confirm the command includes the -- separator.
• If the AI starts a compiled binary rather than source code, ensure dist/backlog has been rebuilt.