Yes — that's the primary use case. Install the Planwright MCP server and Claude Desktop becomes your PM interface: list scheduled objectives, claim them, post plans, record diffs, and request acceptance without leaving your AI client. The same works in Claude Code, Cursor, Codex, and any MCP-compatible client.

Any MCP-capable client works. Add the Planwright MCP server to your client's config, point it at your workspace, and the full agent workflow is available as tool calls. Cursor, Codex, and other MCP-compatible agents can claim objectives, post plans, record diffs, and move work to acceptance — all from inside the editor.

Linear made execution agent-native — you can assign an issue to an agent and it'll write code with Claude Code or Codex. That's real, and it's good. But it accelerated the middle of the pipeline and left the two bookends running at human speed. In Linear, planning is still a person operating the app: opening issues, typing them out, clicking through the UI to feed machines that run thousands of times faster. And acceptance is still the oldest ritual in software — a human opens a pull request and reads it. Those two stages now govern your entire throughput, and Linear didn't touch them. Planwright isn't 'Linear with better agents' — it inverts the two ceremonies Linear left human-paced: Planning happens from inside Claude Desktop, where Planwright synthesizes your chaos — transcripts, decks, email, Slack — into structured objectives with acceptance criteria. The agent drafts from raw signal; you approve. And when objectives complete, Planwright triages them — combing the mechanical criteria, clearing what it safely can, and routing only the judgment calls to you in Cowork. The machine doesn't accept; it sorts. The moment your agents produce more than your team can plan and accept, the human-paced ceremony is the ceiling. Planwright is built so it isn't.

It means the control plane is designed to be driven from inside your AI assistant, not alongside it. Instead of switching to a browser to manage tickets, you ask Claude Desktop to list your scheduled objectives, claim the next one, and check in progress — all via MCP tool calls. The board state and the agent's working context stay in the same place.

It's designed to. Under the 2026 Trust Services Criteria, auditors increasingly expect immutable, tamper-evident audit logs and documented change-approval controls for AI-generated code. Planwright's hash-chained, ECDSA P-256-signed audit records with human acceptance gates map to CC8.1 (change management) and CC7.2 (monitoring). The signed export bundle is the artifact you hand your auditor.

Every agent-generated code change moves to the 'acceptance' lane before it can be marked 'done.' A human engineer reviews the diff, test results, and plan — then signs off. This satisfies HIPAA's requirement for documented review of system changes (§164.312(e)(2)(ii)), and the audit record captures who approved what, when, and with what context.

The audit chain meets NIST 800-53 AU-10 (non-repudiation) and AU-11 (audit record retention) requirements. Each record is signed with ECDSA P-256 via AWS KMS (FIPS 140-2 Level 2). For FedRAMP Moderate authorization, you'd deploy Planwright in a FedRAMP-authorized AWS region — our EU data residency option (Phase 3) extends this to GovCloud.

Linear's agents can close their own issues — there's no enforced human review gate. And while Linear is a mature, widely-adopted product used by tens of thousands of teams, it accelerated execution, not the planning and acceptance boundaries that now govern throughput. Plane publishes its own compliance certifications, but doesn't expose a cryptographic audit artifact that an external auditor can independently verify. Planwright's acceptance-gate state machine and signed hash-chain verification are the differentiators for teams that need both higher throughput and evidence.

A git log records code changes. The audit chain records every decision: who scheduled the objective, which agent claimed it, what plan the agent proposed, what diff was produced, who reviewed it, and who approved it. Each record is hash-chained to the previous one and signed with a dedicated KMS key — so any tampering is detectable. A git log doesn't capture the approval workflow.

Yes. Planwright's acceptance gate ensures a registered representative or qualified person reviews every agent-generated change before it goes live. The audit chain provides the supervisory documentation FINRA examiners expect under Rules 3110 (supervision) and 4370 (business continuity). The signed export bundle can be produced on demand for regulatory review.

Not yet — self-hosted deployment is on the Phase 3 Enterprise roadmap. Today, Planwright runs on dedicated AWS infrastructure (ECS Fargate, RDS, KMS) in a single-tenant account. For air-gapped or IL5 requirements, contact us to discuss a design-partner arrangement for the self-hosted edition.

Free for a single user with 30-day audit retention — once a second person joins, the workspace is a team and moves to a paid plan. Team tier at $15/seat/month includes 1-year audit retention. Business tier at $39/seat/month extends to 3-year retention with priority support. Enterprise tiers (with SSO, SCIM, and data residency on the roadmap) are custom-priced. Discount codes can be entered at checkout. See the pricing section above.