AI Agents (MCP)

YAPL exposes an MCP connector — an integration built on the Model Context Protocol, the open standard AI agents use to work with external tools. Once connected, your agent can read your plans, edit drafts alongside you, record progress, book resources, and answer questions about schedule and cost — always with exactly the access you have, never more.

We design and test the experience with Claude first, but the connector is not Claude-specific: any AI agent that supports remote MCP servers can connect — the protocol, sign-in, and permissions work the same way for every client.

The connector is available on Professional and Trial plans (like the Slack integration); see Capabilities & Security for how the plan check works.

The One Model to Understand: Live vs. Draft

Everything your agent does falls into one of two layers, and knowing the difference explains almost everything about how it behaves:

• Live updates — task status, progress percentages, assignees, actual dates. These apply to your live plan immediately, just like clicking in the app, and they work whether or not you have YAPL open.
• Draft edits — creating or deleting tasks, changing planned dates and durations, dependencies, resource bookings, and publishing. These happen in the plan you have open in your browser, as an unpublished draft. Your published plan is untouched until you review the changes and publish — the same version discipline you already use.

Because drafts live in your browser, structural editing needs a YAPL tab open with the plan on screen. Your agent will tell you when it needs one.

What Your Agent Can Do

The connector exposes 38 tools. You never call them directly — you just talk to your agent — but they group into a few intents:

• Explore and read — list workspaces, projects, teams, and plans; read any plan's tasks, dependencies, and resource assignments.
• Edit the open draft — create, update, move, and delete tasks; build hierarchies; wire dependencies with lags; bulk-import a whole plan outline; undo and redo.
• Track work — set status and progress, log actual dates, record actual work on resource assignments.
• Resources and costs — search your resource pool, book resources onto tasks with correct rate math, and read cost/EVM metrics including your unpublished draft's figures.
• Answer questions — project health, budget usage, schedule metrics, resource utilization.
• Navigate — "take me to the resources page" opens it in your browser tab.

See Working With Your Agent for concrete examples of what these look like in practice.

What It Can Never Do

Your agent acts as you, with your permissions — it can never see or change anything your own account couldn't. On top of that, a set of actions is never exposed to agents at all: account and password changes, billing and subscription management, team membership and role changes, and workspace or plan deletion. Those stay in your hands, in the app.

More detail in Capabilities & Security.

In This Section

• Connecting Your Agent — set up Claude or any other MCP-compatible agent.
• Working With Your Agent — example conversations and what to expect.
• Capabilities & Security — the permission model, in plain words.
• Troubleshooting — the messages you might see and what they mean.