Supported command-line agents
Agent Kanban PM includes adapters for popular coding-agent CLIs and a generic process path for tools that can run in a terminal. The board manages tasks, lifecycle, evidence, and recovery; each selected agent remains responsible for its own code-generation and tool-use behavior.
- Claude Code for Anthropic-powered repository work.
- Codex for OpenAI coding-agent sessions.
- OpenCode for provider-flexible terminal workflows.
- Aider for Git-aware pair programming.
- Antigravity CLI and custom shell adapters for additional runtimes.
Four ways to connect
Managed local sessions
Choose an agent on a task and launch it from the board. Agent Kanban PM starts the configured executable, streams terminal output, records process state, and keeps the session associated with the card. Optional worktree isolation gives the run a clean branch and working directory.
MCP tools
Agents that support the Model Context Protocol can interact with project and task state through Agent Kanban PM's MCP server. This is useful when an orchestrator needs to create work, claim a task, post progress, attach evidence, or request a stage transition using structured tools.
REST API
The local HTTP API exposes the same core domain for scripts and purpose-built integrations. Use the generated API schema and the local authentication token rather than scraping the browser interface.
Command-line interface
The CLI supports terminal-first operators and automation. It is also the simplest bridge for shell scripts that need to inspect or update board state without embedding an HTTP client.
Integration boundaries
Installing Agent Kanban PM does not install or subscribe you to third-party agent services. Install and authenticate each coding-agent CLI according to its provider's instructions, then configure the executable locally. Provider charges, if any, are separate from this project.
Agent Kanban PM has no paid plan, per-agent charge, hosted account, or usage fee. It is open source under the MIT license and runs locally.
A shared contract across different agents
Using several agents is manageable when they share an operational contract. The task holds the goal and acceptance criteria. Claims and leases establish ownership. Events expose progress. Review evidence connects the run to files, tests, and commits. Approval gates keep sensitive transitions under human control.
That contract is intentionally independent of the prompt syntax or model vendor. You can change the assigned CLI without replacing the project-management layer or losing the task's history.
Start with one agent
You do not need a multi-agent setup on day one. Start the local board, connect the coding CLI you already use, and run one well-scoped task through review. Add orchestrators, MCP clients, or concurrent worktrees after the basic loop is comfortable.