MCP agent workflows

Let your AI do the writingCopy Anchor Link

Your AI can write notes, flashcards, practice tests, task lists and timetables itself and save them to WorkRoad. Ask it, for example, "Make flashcards from yesterday's OS lecture": it reads the lecture's transcript, writes the cards and saves them as a study set. When your AI writes the content itself, it uses none of your WorkRoad AI uses. Only asking WorkRoad's AI to write something, or using Copilot, counts. WorkRoad asks connected AIs to work this way, and the tools that run WorkRoad's AI say that they use an AI use, so your AI calls them only when you ask for WorkRoad's AI. The exception is grading: WorkRoad's AI grades the short answers in a practice test when you submit it. In a test your AI wrote, that counts one AI use per submit; in a test WorkRoad's AI wrote, it is included.

In Claude Code, /mcp__workroad__make-cards <note or lecture> does it in one step. For developers, Let your AI do the writing lists the operations.

PromptsCopy Anchor Link

The server also offers prompts. Claude Code lists them as slash commands:

Read the docs from an agentCopy Anchor Link

An agent can read these docs to answer questions about WorkRoad itself: how a feature works, what a plan includes, or how to call the API. The server's instructions tell it to look the answer up rather than guess.

• searchDocs finds the pages that match a few words, with the passages that matched, and getDoc returns a page as Markdown. Both take a locale for a translation, and any key or connected app may call them.
• The server also lists every page as a resource, at the page's Markdown address, so a client such as Claude Code can attach one to a conversation.
• Without the MCP server, add .md to a docs address for the page as Markdown, as in https://workroad.app/docs/api/mcp.md, or ask for the page with Accept: text/markdown. Add ?lang=de, es, fr or it for a translation.
• /llms.txt lists every page by section, and /llms-full.txt has them all in one file.
• On each docs page, Copy page copies it as Markdown, and Open opens it in ChatGPT or Claude.

Practising from an agentCopy Anchor Link

A practice session records real reviews, the same as the app's Review page:

1. getToday returns what is due, and a practice suggestion with the arguments for the next call.
2. startPracticeSession, optionally with subjectId or setIds, returns the first cards.
3. For each card the agent shows the term, waits for the answer, reveals the definition and calls answerPracticeCard with the card's itemId, a rating (again, hard, good or easy) and durationMs.
4. When remaining reaches 0, getPracticeSession refills the session. finished: true means nothing is due now.

undoPracticeAnswer takes back the newest answer. Practice sessions has the details.

Long-running workCopy Anchor Link

Tools that start AI work or parse a file, such as generateNoteFromTopic, sendCopilotMessage or completeNoteFile, return a jobId straight away. Poll getJob with that id as jobId until status is succeeded, failed or cancelled, waiting the retryAfter seconds from the text block between calls. Jobs describes each kind's result.

The webhook tools appear only for an API key with webhooks:manage, which a connected app never gets. Creating a webhook, changing its URL and rotating its secret are not tools: a webhook sends events to the URL it was given for as long as it exists, and an agent steered by text it read could point it at someone else's server. Do those in the app or over REST.