Planwright

Control plane for autonomous software labor.

Planwright is the objective-native planning board for agent-led software development. Humans define outcomes at the Objective layer; coding agents (Claude Code, Cursor, Codex, and others) claim work, decompose tasks, execute, and report back via native MCP—with every state change captured in a hash-chained, Ed25519-signed audit trail.

Quick Start

Claude Code (recommended)

Terminal

claude mcp add planwright https://mcp.planwright.tools/mcp

Your browser opens for GitHub login. That's it.

Other Agents

See the full agent setup guides for Cursor, Codex, Claude Desktop, and generic MCP clients.

How It Works

Code

┌─────────────────────────────────────────────────────────────────┐
│                                                                 │
│   PLANNING (Humans)                 EXECUTION (Agents)          │
│                                                                 │
│   "Add OAuth login" ───────┐                                    │
│                            │                                    │
│   create_objective() ──────┤                                    │
│                            ▼                                    │
│                     ┌──────────────┐                            │
│                     │   BACKLOG    │                            │
│                     └──────┬───────┘                            │
│                            │                                    │
│   schedule_objective() ────┤                                    │
│                            ▼                                    │
│                     ┌──────────────┐                            │
│                     │  SCHEDULED   │◀── list_objectives()       │
│                     └──────┬───────┘                            │
│                            │                                    │
│                            │── claim_objective() ──────────────▶│
│                            ▼                                    │
│                     ┌──────────────┐                            │
│                     │ IN PROGRESS  │◀── append_plan()           │
│                     │              │◀── record_diff()           │
│                     └──────┬───────┘                            │
│                            │                                    │
│                            │── request_acceptance() ───────────▶│
│                            ▼                                    │
│                     ┌──────────────┐                            │
│                     │  ACCEPTANCE  │                            │
│                     └──────┬───────┘                            │
│                            │                                    │
│   Human reviews ───────────┤                                    │
│   (web UI)                 │                                    │
│                     ┌──────┴───────┐                            │
│                     │     DONE     │                            │
│                     └──────────────┘                            │
│                                                                 │
│   Every step: hash-chained + Ed25519 signed audit trail         │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Key Features

Objective-Native, Not Task-Based No story points. No velocity charts. No capacity planning. Agents decompose objectives into tasks themselves—you manage outcomes, not effort.

Immutable Audit Trail Every state change produces a hash-chained, Ed25519-signed audit record. Know exactly what the agent planned, what it changed, and when.

Native MCP Integration Agents connect via the Model Context Protocol—no custom API clients needed. Works with any MCP-compatible agent.

Multi-Agent Support Claude Code, Cursor, Codex, Factory, Amp—if it speaks MCP, it can work from your Planwright board.

Context Files Push CLAUDE.md, architecture docs, and specs to Planwright. Agents read them before starting work.

Documentation

Examples

See the examples/ directory for:

Sample CLAUDE.md files for different project types
Example objective structures
Context file templates

MCP Server

Planwright's MCP server is hosted and managed—no self-hosting required.

Endpoint	Protocol
`https://mcp.planwright.tools/mcp`	Streamable HTTP

Authentication is handled via OAuth 2.1 with Dynamic Client Registration (RFC 7591). Your agent opens a browser for GitHub login on first connection.

For CI/CD and headless environments, generate a static token in Settings → MCP Tokens.

Available Tools

Planwright exposes 30 MCP tools for agent orchestration:

Setup & Context

Tool	Description
`planwright_set_repo`	Set active project for the session (by GitHub repo or project ID)
`planwright_list_workspaces`	List workspaces you have access to
`planwright_list_projects`	List projects with IDs and linked GitHub repos
`planwright_list_context_files`	List CLAUDE.md, specs, architecture docs for the project
`planwright_get_context_file`	Read a specific context file
`planwright_push_context_file`	Upload or update a context file
`planwright_get_doc_suggestions`	Get pending documentation gap suggestions
`planwright_dismiss_doc_suggestion`	Dismiss a doc suggestion
`planwright_get_board_url`	Get the Planwright board URL

Initiatives

Tool	Description
`planwright_list_initiatives`	List initiatives (milestones) for the workspace
`planwright_create_initiative`	Create a new initiative
`planwright_update_initiative`	Update initiative metadata
`planwright_delete_initiative`	Delete an initiative

Objectives

Tool	Description
`planwright_list_objectives`	List objectives by lane (backlog, scheduled, in_progress, acceptance, done)
`planwright_get_objective`	Get full objective details including agent run history
`planwright_create_objective`	Create a new objective with acceptance criteria
`planwright_update_objective`	Update objective metadata
`planwright_schedule_objective`	Move objective from backlog to scheduled

Agent Workflow

Tool	Description
`planwright_claim_objective`	Claim a scheduled objective and start an agent run
`planwright_append_plan`	Post the agent's decomposition plan
`planwright_check_alignment`	Verify work aligns with claimed objective (drift detection)
`planwright_record_diff`	Record diff summary and test results
`planwright_submit_test_run`	Submit acceptance criteria verification results
`planwright_request_acceptance`	Move objective to acceptance lane for human review
`planwright_append_note`	Record task completion in the agent run log
`planwright_flag_clarity`	Flag objective as having clarity concerns
`planwright_override_clarity`	Human override for clarity flags

Enrichment & Classification

Tool	Description
`planwright_list_unenriched`	List objectives/initiatives missing intent summaries
`planwright_list_unclassified`	List objectives missing strategic classification
`planwright_list_my_bugs`	List bugs for this project

Support

License

Planwright is a commercial product. The MCP integration examples in this repository are provided under the MIT License.

See LICENSE for details.