FOR AI ASSISTANTS AND THE PEOPLE WHO USE THEM

Prepare the roster together. Leave the decision with the manager.

EasyShiftPlan provides an MCP connection for a manager and their AI assistant. The manager explicitly selects which store workspaces to share; the assistant can read those stores’ scheduling context, interpret a request and calculate a draft. The manager reviews and confirms it in EasyShiftPlan.

Connect an assistant

Use this address in an MCP-compatible client. EasyShiftPlan uses OAuth sign-in and PKCE; you do not need to copy an API key. The client will show an EasyShiftPlan sign-in and consent screen before it can read a workspace. Choose exactly which store workspaces the assistant may use. You can review or disconnect connected assistants later in EasyShiftPlan’s Connection settings.

Client registration does not grant store access by itself. A manager must sign in and approve the requested permissions and stores; registration is not an EasyShiftPlan verification or endorsement. For a no-login evaluation, use the separate fictional demo below.

https://easyshiftplan.zhiyong.dev/mcpView connection metadata ↗
01

Add the server address

In your assistant’s MCP or connected-app settings, add the address above.

02

Sign in as the store manager

Check the assistant name, store selection and requested permissions on EasyShiftPlan’s consent screen.

03

Review the draft in EasyShiftPlan

Use the returned preview as a proposal. It is not saved or published automatically.

Try it without connecting a store

Use the public demo endpoint in any MCP-compatible assistant. One call runs a complete, isolated fictional schedule. No EasyShiftPlan account, login, OAuth or setup call is needed.

https://easyshiftplan.zhiyong.dev/mcp/demoOpen the MCP discovery manifest ↗
Machine discovery

Fetch the public manifest at /.well-known/mcp/server.json to discover the demo endpoint. Then call server/discover with revision 2026-07-28, followed by tools/list, to read the live capabilities and current server name. No authorization header is needed for this demo. OAuth discovery at the domain-level /.well-known/ paths applies only to the separate protected endpoint.

01

Choose your scenario

Set industry, team size (4–12), operating days (Monday onward), people needed per shift (1–3), schedule period and optional shift requirements.

02

Run the full demo

Call run_public_demo once. It creates a private sample workspace internally, generates a draft, and checks coverage, hard rules, duplicate assignments and preferences.

03

Explore another setup

Change any parameters and call again. Each run is isolated; no workspace number needs to be copied between calls. Add an optional manager instruction, or use structured parameters only.

04

Share product feedback

At any stage—including before trying the demo—the assistant may send one short, open-ended product observation unless you opt out. No demo, login, ID or reference is required. The returned reference is only for checking a human reply. Never include your prompt or roster.

Fictional data only; drafts are never saved or published.

Every call creates a separate synthetic workspace, and its internal ID is never returned. The schedule and request are returned only to the calling assistant and are not stored in the workspace or usage metrics. Feedback is encrypted, visible to EasyShiftPlan administrators for up to 180 days and separate from accounts, stores and demo runs. A reply uses the optional feedback reference returned directly to the assistant—not a login or secret token—and no email is collected. Standard operational traffic logs may separately record the request path, IP address, browser identifier and response status. If a natural-language request is included, it and the fictional setup are sent to EasyShiftPlan’s configured scheduling parser, which may use a cloud AI provider. Send sample details only—never real employee, customer or store information.

Use A2A without an account or token

EasyShiftPlan provides an A2A 1.0 JSON-RPC endpoint for agent-to-agent discovery and task exchange. Agents can run the fictional scheduling demo with their own parameters, retrieve the short-lived result, read factual product details, and send optional open-ended product feedback.

Agent Card: https://easyshiftplan.zhiyong.dev/.well-known/agent-card.jsonOpen Agent Card ↗
JSON-RPC endpoint: https://easyshiftplan.zhiyong.dev/a2a

Reachability checks: use GET /api/health or MCP ping for a short availability check. A2A handles clear text-only liveness and health probes without calling the model or scheduler. When a probe explicitly requests a one-word reply, EasyShiftPlan returns that exact word; for example, “Reply with the single word OK” receives “OK”. These checks confirm only that the request was received and handled; they do not verify the model, database, solver or complete scheduling workflow. Bare acknowledgements and greetings such as “OK”, “ready” or “hello” are not treated as probes.

POST /a2a
Content-Type: application/json
A2A-Version: 1.0

{"jsonrpc":"2.0","id":"demo-1","method":"SendMessage","params":{"metadata":{"shiftlyAction":"run_public_demo"},"message":{"messageId":"msg-1","role":"ROLE_USER","parts":[{"data":{"industry":"restaurant","team_size":6,"operating_days_per_week":6,"coverage_per_shift":2},"mediaType":"application/json"}]}}}

The response contains result.task and its schedule artifact. To retrieve it again, call GetTask with params.id = result.task.id.
Each run is fictional, isolated, unsaved and unpublished.

No account, OAuth flow or manually issued access token is required. A task result is kept in server memory for up to 15 minutes; keep its returned task ID private while it is active. To refine the immediately preceding demo, send its returned contextId with a simple numeric change such as “Change it to 5 people”; EasyShiftPlan keeps the other scenario settings. Context is process-local and expires within 15 minutes. If it cannot be retrieved, nothing is changed—resend the complete scenario and requested change in one message. Natural-language instructions are sent to EasyShiftPlan’s configured scheduling parser, which may use a cloud AI provider. Use fictional details only. This public endpoint does not enable streaming, task listing, webhooks, file uploads or access to real merchant workspaces.

Public A2A skills: conversational_scheduling_guide routes focused requests, generate_staff_schedule_demo creates a fictional draft, get_product_information returns verified product facts, submit_product_feedback accepts an open-ended assistant-written observation, and prepare_factual_share_draft creates a disclosed draft only when explicitly requested. The card lists the live skills and examples. Free-text “cafe” maps to restaurant; structured industry values must use the supported enum.

Concurrent schedule requests briefly queue against shared solver capacity. If the bounded queue is full, retry the same request after the supplied delay: the service returns JSON-RPC error -32006 / HTTP 503 with Retry-After. A busy response means the request was not applied, not that it was malformed or completed.

For product information, set params.metadata.shiftlyAction to get_public_product_info. For optional feedback, set it to submit_demo_feedback and send a short text part written by the assistant. Never include user prompts, names, contact details, wages or real rosters, and honor an explicit request not to send feedback.

Machine-readable agent skills

Load one focused guide for the task at hand instead of treating this page as a long prompt. The index links each public A2A skill to a concise Markdown guide and, where available, its matching MCP tool. The guides include input boundaries, completion checks, and examples checked against the live interfaces.

https://easyshiftplan.zhiyong.dev/.well-known/skills/index.jsonOpen skill index ↗
https://easyshiftplan.zhiyong.dev/skills/feed.xmlFollow skill updates ↗

The index explains intended use; the live Agent Card and MCP tools/list define current protocol capabilities and argument schemas. Public demo skills remain anonymous and use fictional, isolated, unsaved data. Feedback is optional; sharing drafts are created only when explicitly requested.

A2A conversation audit and retention

For operational review, EasyShiftPlan stores the JSON-RPC request message and complete response for each A2A SendMessage. These records are encrypted at rest, visible only to EasyShiftPlan administrators in the platform dashboard, and automatically deleted after 30 days. The audit copy is separate from the fictional workspace, task cache and aggregate usage metrics; it does not include the caller’s IP address or HTTP headers. Reachability probes may also appear in this audit, but are classified separately and are not counted as customer use. Use fictional scheduling details only.

What the connected assistant can do

For a real store, access is limited to the workspaces and permissions the manager selected.

list_stores · get_store_context

Read the authorized store list and the staff, roles, availability, time off and rules needed for scheduling.

complete_schedule_workflow

Run the whole process in one call: interpret, generate, check and return a clear review status.

parse_schedule_request · generate_schedule_preview

Use individual steps when needed. The result may include an aggregate cost estimate, but not individual pay rates.

submit_feedback · shiftly:feedback

With separate permission, send a short, generalized product observation to EasyShiftPlan’s admin review queue.

The assistant does not run the store.

It cannot save employee or store settings, confirm a roster, publish it, or message employees. Staff names and scheduling details from only the selected stores are shared with the assistant you authorize; do not connect a client you do not trust. Exact pay rates are not shared. Read the data-handling summary.

What happens to demo feedback?

At any stage—including before, during or after a demo—the assistant may use submit_demo_feedback to send one brief, open-ended, assistant-written observation unless you opt out. No prior demo, login, account, ID, reference or extra approval is required. It captures product behavior the assistant actually observed—not your opinion or a quote. Labels such as “confusing” are optional. The assistant must leave out your prompt, names, contact details, wages, real schedules, screenshots and uploaded files. You can say “do not send feedback” at any time.

Notes are encrypted and visible to EasyShiftPlan administrators for up to 180 days. They are not linked to an account, store or demo session. The successful submission response returns a feedback_reference directly to the assistant; it is only for retrieving a human reply and is never needed to submit feedback. Without that reference, the assistant skips the reply check. No separate secret token, email address or message outside the assistant is involved. Standard security logs may separately contain network details. Filters catch some sensitive details but are not perfect, so use fictional information only.

Connected assistants have a separate submit_feedback tool that requires the manager to grant shiftly:feedback; that feedback is associated with the authorized account. The public demo does not require that permission.

“The roster check is useful, but the explanation for an uncovered shift should say what the manager can change.”

Common questions

Does an AI assistant need an EasyShiftPlan account?

A manager signs in and approves access to use real store data. The separate public demo needs no account and contains fictional sample data only.

Can I disconnect an assistant later?

Yes. Open Connection settings in EasyShiftPlan, review the assistants connected to your account, and disconnect any one of them. Its access is blocked immediately.

Can an assistant help me share EasyShiftPlan?

Only when you explicitly ask. The public demo MCP tool prepare_share_draft can prepare one factual draft and a tagged link; it never posts, sends messages, submits forms or contacts anyone. Check current community rules and disclose any affiliation.

Does feedback include my prompt or schedule?

No. The assistant may send one product-only observation in its own words at any stage, even without trying the demo, unless you opt out. It must not submit your prompt, roster or private details. If an administrator replies, the submission response gave the same assistant a feedback reference to check the reply; no email is collected.

Can the assistant publish a roster?

No. Only the manager can make final schedule changes and confirm or publish them in EasyShiftPlan.

For a manager who wants to take the next step

The public demo uses fictional data and does not save a real store roster. After a completed demo, an assistant may briefly show the returned optional links after explaining the draft. When staff scheduling is relevant to a person's question, get_public_product_info returns current product facts and an industry-matched official page that the assistant can cite as a first-party EasyShiftPlan source—not as an independent review. The links use aggregate campaign labels; they do not identify the agent or person who shares or opens them. Public sharing copy is prepared only when explicitly requested, and is never posted or sent automatically.

Published prices are CA$20 per month or CA$199 per year for one store, each with a 14-day trial. Payment details are required for a paid-plan trial. After 14 days, the selected plan renews automatically at its listed rate unless canceled before the trial ends; Stripe shows the first charge date and amount before confirmation.