ClassFlowy connector for AI assistants (MCP)
Last updated: 29 September 2026 · Wersja polska: Ustawienia → Chatboty w aplikacji
ClassFlowy is a school notebook for students: it turns lesson recordings and photos of the board into lessons in a school calendar, with AI-generated lesson summaries, notes and a weekly timetable. The ClassFlowy connector lets an AI assistant you choose – Claude, ChatGPT, Gemini or any client that supports the Model Context Protocol – read your lessons and help you study, with your permission.
Server addresses
https://classflowy.com/api/mcp/directory– the version listed in the Claude and ChatGPT directories. Every tool calls only ClassFlowy's own services, so web search is not included.https://classflowy.com/api/mcp– the full version for connecting ClassFlowy yourself as a custom connector. It adds web_search.
Both are remote MCP servers over Streamable HTTP. Authentication is OAuth 2.1 with PKCE: your assistant opens ClassFlowy's sign-in page and asks for your consent. You never paste a password or a key into the assistant. An authorization granted for the directory version does not open the full version.
Requirements
- A ClassFlowy account with a confirmed email address.
- A plan that includes MCP calls: Plus, Pro, Max – or an MCP add-on on any plan.
- An AI assistant that supports remote MCP servers with OAuth.
How to connect
- Claude (claude.ai, desktop and mobile apps): Settings → Connectors → Add custom connector → paste the full server address → Connect → sign in to ClassFlowy → Allow. (If ClassFlowy is listed in Claude's connector directory, you can also add it from there.)
- ChatGPT: Settings → Apps & Connectors → Advanced → turn on developer mode → Create → paste the server address, authentication OAuth → sign in to ClassFlowy → Allow.
- Gemini (gemini.google.com, where custom apps are available): Settings → Connected Apps → Add a custom app → paste the server address. Gemini Enterprise needs a client ID and secret: create them in ClassFlowy → Settings → Chatbots → OAuth client, using the redirect URI shown by Gemini.
- Claude Code, Cursor and other MCP clients: add an HTTP MCP server with the server address; the OAuth sign-in starts automatically.
What the assistant can do
| Tool | What it does | Access | Directory version |
|---|---|---|---|
get_calendar | Returns one Monday–Sunday week of the student's school calendar that contains the given date (default: the current week): lessons with uploaded materials (subject, time, topic, number of materials and notes) and timetable slots that have no materials yet. Lesson content is usually in Polish. | read | yes |
list_lessons | Lists the student's lessons, newest first, optionally within a date range, for one subject (by name), and/or matching keywords. Keywords are matched against lesson summaries and notes only; recording transcripts are not searched. Returns lesson ids for get_lesson. | read | yes |
get_lesson | Returns one lesson: its AI-generated summary (Markdown), topic and date, notes added by the student or by an assistant, and the list of attached materials (type and title only). Transcripts and OCR text of individual materials are not available through this connector. | read | yes |
get_timetable | Returns the student's weekly timetable: the bell schedule (lesson number to start and end time) and timetable entries (weekday, lesson number, subject, room, week parity). | read | yes |
list_subjects | Lists the student's school subjects, including subjects classmates have shared with them, with the number of lessons and the date of the latest lesson. | read | yes |
add_lesson_note | Appends a note to a lesson, labelled in ClassFlowy as a note from an assistant. The note is shown in the lesson and is taken into account the next time the lesson summary is generated. Existing notes are not changed. | write | yes |
knowledge_search | Searches the knowledge base the student keeps in ClassFlowy: named entries such as goals, exams, deadlines, people or preferences, each with short observations and relations to other entries. This is ClassFlowy's own data, separate from the AI client's memory. | read | yes |
knowledge_open | Returns the named entries of the student's ClassFlowy knowledge base with all their observations (with ids) and relations. | read | yes |
knowledge_save | Saves facts to the student's ClassFlowy knowledge base. An existing entry with the same name is extended with the new observations instead of being duplicated. Each observation is one short, self-contained sentence. | write | yes |
knowledge_link | Creates relations between entries of the student's ClassFlowy knowledge base, for example from 'Student' via 'is preparing for' to 'Physics final exam'. Entries that do not exist yet are created. | write | yes |
knowledge_delete | Permanently deletes a whole entry of the student's ClassFlowy knowledge base (by name) or a single observation (by an observationId returned from knowledge_open or knowledge_search). | write | yes |
web_search | Searches the public web through ClassFlowy's search provider (Tavily) and returns page titles, URLs and text snippets from third-party websites. Each search uses the student's ClassFlowy web-search allowance. | read | no – custom connector only |
run_python | Runs a Python 3 program in ClassFlowy's isolated sandbox (no network access; scientific libraries such as numpy and sympy when installed) and returns its standard output, errors and result. Images and files created by the program are not returned through this connector. Each run uses the student's ClassFlowy sandbox allowance. | read | yes |
On the consent screen you can untick write access; the assistant then sees only the read tools. Transcripts and OCR text of individual recordings and photos, original files and your conversations with the ClassFlowy assistant are never available through the connector.
Example prompts
- “What did we cover in my last physics lesson? Summarize it in five bullet points.” (list_lessons → get_lesson)
- “What's on my timetable tomorrow, and which of this week's lessons already have notes?” (get_timetable, get_calendar)
- “Add a note to today's history lesson: essay on the causes of World War I is due on Friday.” (list_lessons → add_lesson_note)
- “Remember that my maths final exam is on 12 May and that trigonometry is my weakest topic.” (knowledge_save)
- “Make me a three-day revision plan for my biology test based on my last four biology lessons.” (list_lessons → get_lesson)
Limits
Every tool call uses one MCP call from your monthly allowance – including calls that fail.
- Standard: no MCP access
- Plus: 80 MCP calls per month
- Pro: 1,000 MCP calls per month
- Max: 10,000 MCP calls per month
Add-ons: 100 calls for 5 PLN, 200 calls for 10 PLN, 500 calls for 20 PLN, 10,000 calls for 50 PLN. Web search and Python additionally use the plan's web-search and sandbox allowances, exactly as in the ClassFlowy assistant.
Data and privacy
- Data leaves ClassFlowy only when your assistant calls a tool during your conversation with it, and only after you allowed the connection.
- ClassFlowy does not receive your conversation with the assistant – only the tool calls, which count towards your plan.
- Access tokens expire after one hour, refresh tokens 30 days after their last renewal. ClassFlowy stores all tokens only as SHA-256 hashes.
- You can disconnect an assistant at any time in ClassFlowy → Settings → Chatbots; access ends immediately.
- Web search (full version only) sends the query to ClassFlowy's search provider, Tavily. Python runs on ClassFlowy's own server without network access.
Full privacy policy (Polish): /prywatnosc · Terms of service (Polish): /regulamin · Payment terms (Polish): /regulamin-platnosci
Support and security
Questions and support: system@biskupski.site. Security issues: the same address, see security.txt.