# MCP Connector for AI Agents: Let Your Agent Run the Plan | YAPL | YAPL (https://yapl.app/en/mcp-integration) > Connect Claude, ChatGPT, Gemini, or any MCP-compatible AI agent to YAPL. Agents draft tasks and dependencies, book resources with real rate math, and answer cost and cash flow questions. Every structural edit lands in a draft you review and publish. **Your AI agent, hands on the schedule** YAPL speaks the Model Context Protocol, the open standard AI agents use to work with real tools. Connect the agent you already talk to every day and let it plan with you: drafting tasks, wiring dependencies, booking resources, and answering cost questions inside your actual plan. ## Works with the agents you already use One connector URL, the open MCP standard, and OAuth sign-in. No API keys, nothing to install. Claude · Claude Code · ChatGPT · Codex · Gemini CLI · Antigravity · Cursor · Windsurf · OpenClaw *All product names and trademarks are the property of their respective owners, mentioned for interoperability only; no endorsement is implied.* ## One open connector, not a bolted-on chatbot Most AI features are a proprietary assistant living in someone else's sidebar. YAPL took the opposite path and exposes a standard MCP server that any capable agent can connect to. You choose the agent. The same one that reads your email, writes your code, or sits in your terminal now gains hands for project planning. Connection is self-serve: add the connector URL, sign in with your YAPL account in your own browser, and approve access. The agent acts strictly as you, with your workspace permissions, and you can revoke any client at any time from settings. ## Project health in plain words Agents read the same live metrics your dashboard shows, from progress and milestones to budget usage and earned value indices, and answer like a colleague rather than a report. No tab open, no export, no query language. The question you would ask a project manager is the whole interface. > **You:** How is the Riverside project doing? Are we on budget? > **Agent:** Riverside is 63% complete. Two of five milestones are done and nothing is overdue. Earned value is running ahead of spend with a CPI of 1.29, so you are under budget for the work performed. ## Resource booking that knows your prices Say the material and the amount in your own words. The agent searches your resource pool, reads the current rate, checks availability, and books the assignment. Planned work and cost come from YAPL's own engine, never from the model's imagination. > **You:** Put 65 cubic meters of C25 concrete on the Concrete Pour task. > **Agent:** Booked 65 m³ of Ready Mix Concrete C25 on Concrete Pour. Planned cost is $6,175 at the current $95/m³ rate. ## Cost and cash flow, before you commit The agent computes earned value metrics and weekly cash flow on your unpublished draft, using the exact same math as your cost dashboard, and compares it with the live plan. You see what publishing would change while it is still just a draft. > **You:** What does this draft do to our cash flow? > **Agent:** The draft adds $950 to week 31, moving budget at completion from $127,600 to $128,550. Cost performance stays healthy at a CPI of 1.05. ## Powerful hands, careful by design Structural changes such as new tasks, dependencies, planned dates, and resource bookings land in an unpublished draft in your open plan, exactly as if you had made them yourself. You watch edits appear live, the diff shows every change, undo works, and nothing touches the published plan until you review and publish. Runtime updates like status and progress apply immediately, the way a teammate clicking in the app would. A fixed set of actions is never exposed to agents at all: account changes, billing, member and role management, and deleting plans or workspaces stay in your hands. ## Connected in a minute No API keys, no configuration on YAPL's side. 1. **Add the connector** — In your agent's connector or MCP settings, add https://yapl.app/api/mcp as a remote MCP server. 2. **Sign in and approve** — Your browser opens YAPL's sign-in page. Approve the client and it receives your basic identity and expiring tokens, never your password. 3. **Talk to your plan** — Ask for a draft, a booking, or a budget answer. Try: "List my plans." ## The rules every agent plays by - **Acts as you, sees as you** — Every read and write runs under your account with your workspace roles and row-level permissions, and never more. - **Draft-first editing** — Structural edits wait in a reviewable draft. Publishing is always a deliberate human step. - **Revocable per client** — Every connected agent is listed under Settings → Integrations → MCP Connectors, and one click disconnects it. - **Professional & Trial plans** — The connector is checked on every call against your workspace's current plan, just like the Slack integration. ## Frequently asked questions ### Which AI agents work with YAPL? Any agent that supports remote MCP servers with OAuth sign-in. That includes Claude on claude.ai, Claude Desktop, and Claude Code, plus ChatGPT and Codex, Gemini CLI, Google Antigravity, Cursor, Windsurf, and other MCP-compatible clients. The connector follows the open standard with no client-specific extensions, so new agents work without any changes on YAPL's side. ### Which plans include the MCP connector? Professional and Trial plans, the same as the Slack integration. The check runs on every agent call against the workspace's current plan, so a downgrade or trial expiry pauses agent access immediately. Upgrading restores it without reconnecting. ### Can an agent break my published plan? No. Structural changes land in an unpublished draft in your open plan. You watch them appear, review the diff, and only you decide to publish. Undo works throughout, and if the live plan changed while drafting, YAPL's normal review dialog takes over rather than letting the agent merge anything on its own. ### What data does a connected agent see? Exactly what your own account can see: plans, tasks, resources, and metrics in your workspaces, and never more. What the agent reads is transmitted to that agent's operator under their own terms, which is why connecting is an explicit OAuth approval you can revoke at any time. ### Why do some actions need my browser open? Plan drafts live in your browser. That local-first design is what keeps editing instant and your published plan safe. Structural editing therefore needs a YAPL tab with the plan on screen, while reads, metrics, and status updates work with no tab at all. ### Can the agent publish without asking me? Publishing only happens when you tell the agent to publish, and any conflict with live changes stops and hands control back to you. Your version discipline stays intact: the agent drafts, you decide. ### How do I disconnect an agent? Settings → Integrations → MCP Connectors lists every client you've approved, with its connection time and client ID. Revoke disconnects that client immediately, independently of the others. --- **Ready for High Efficiency?** Try before subscribe. Plan your workflow efficiently with YAPL - [Start Free Trial](https://yapl.app/en/register) — 14-Day Trial, No credit card required - [Documentation](https://yapl.app/en/docs)