Developers
Language

Your membership platform,one conversation away

The Cademí MCP connects Claude and other AI assistants to your account. Ask why a student can't get in, what a course contains, or change what needs changing, with your permissions and your name on every action.

Release 0.3.29

One connection.
The whole account.

Almost every operation of API v3 is a tool, generated from the same contract as the API reference. On top of them, composite tools answer whole questions and carry out whole tasks in a single call.

tools
442
composite tools
50
resources
18
error codes
13
account_domains_certificate_getbanners_copies_createcomments_deletedeliveries_copies_createaccount_getbanners_getcomments_replies_createdeliveries_listaccount_usagebanners_updatecomments_replies_updatedeliveries_rules_listadministrators_getcustom_code_getdepartments_deletedeliveries_tags_listimports_analyses_creatediamonds_memberships_createlegal_terms_listcapabilities_getfiles_listfaqs_getproducts_access_schedules_get
audit_entries_getcertificates_createconfiguration_applications_creategateways_getevent_streams_deletecertificates_listcustom_fields_createsales_events_listevents_getcertificates_template_updatecustom_fields_listsales_events_processing_attempts_listexports_deletediamonds_deletelegal_terms_acceptances_listadministrators_permissions_getfiles_deletedepartments_updateenrollments_listimports_getdiamonds_memberships_processing_attempts_createsettings_admin_area_update

Whole tasks, not endpoints

Composite tools answer a whole support question, or carry out a whole task and check the result, in one call. Name people and courses the way you would: the tool finds them, and asks back when a name matches more than one. Tools that change something say so, and your client asks you first.

  1. student_access_context

    Why a student can or cannot get in

    The access verdict, the enrollments for that product, the release calendar and the progress, with a diagnosis in reason codes, the evidence behind it and the next step.

    student_access_contextRead-only
    “Why can't João see module 3?”
    Access
    Partial
    Reason
    CONTENT_PARTIALLY_RELEASED
    Evidence
    Module 3 releases 14 days after enrollment, on October 2
    Next step
    Wait for the release, or grant an exception
  2. investigate_sale

    From purchase to access

    Follows one purchase from the payment through the sales events and their processing to the delivery, the enrollment and the access it produced.

    investigate_saleRead-only
    “Maria paid. Why can't she get in?”
    Payment
    Approved on October 4
    Sales event
    Received, not processed
    Delivery
    None covers offer MKT-ANNUAL
    Access
    Not granted
  3. grant_student_access

    Grant access, and check it

    Enrolls a student in a delivery, applies per-product exceptions when you ask for them, and reads the effective access back, so the answer says what the student can open.

    grant_student_accessAsks before it changes
    “Give João the October package.”
    Student
    João Pereira
    Delivery
    October package
    Access
    3 products, verified
    Retry
    Same request key, applied once
  4. create_delivery

    A delivery in one step

    Creates a complete delivery, with its gateway, offer codes, products, access durations, release schedules and entry tags, after checking what the gateway supports.

    create_deliveryAsks before it changes
    “Create a one-year delivery for Marketing Essentials.”
    Products
    Marketing Essentials
    Duration
    365 days
    Release
    Monthly schedule
    Status
    Created
  5. create_course_outline

    A course outline, as drafts

    Builds a new course in a showcase from module and lesson names: the product, its modules and its lessons are created as drafts, ready for you to write.

    create_course_outlineAsks before it changes
    “Draft a course called Instagram for Creators with 4 modules.”
    Product
    Instagram for Creators · draft
    Modules
    4 · draft
    Lessons
    16 outlines · draft
    Next step
    Write the lessons
  6. support_question_queue

    Questions waiting for an answer

    The student questions still open, by product, student or lesson, so the assistant can draft the answers and publish the ones you approve.

    support_question_queueRead-only
    “Which questions are still unanswered?”
    Open
    7 questions
    Oldest
    Lesson 3.2 · 4 days
    Product
    Marketing Essentials
    Next step
    Answer with support_publish_question_answer
student_access_contextRead-only
“Why can't João see module 3?”
Access
Partial
Reason
CONTENT_PARTIALLY_RELEASED
Evidence
Module 3 releases 14 days after enrollment, on October 2
Next step
Wait for the release, or grant an exception

Ask it like you would ask a colleague

Copy a prompt to try it in your assistant.

  • Diagnose access

    Why can't ana.souza@acme.com access the Marketing course? Tell me what to do about it.

  • Investigate a sale

    Maria Lima paid for the Marketing course yesterday and still can't access it. Find out what happened.

  • Grant access

    Give joao@acme.com the October package. Ask me before you do it.

  • Draft a course

    Create a draft course called Instagram for Creators, with 4 modules and 4 lesson outlines in each.

  • Answer questions

    Which student questions are still waiting for an answer? Draft a reply to the oldest one.

  • Audit a student

    Audit Maria Lima's experience: her enrollments, her progress and anything that blocks her.

Built for the admin, not around them

The assistant works with the same API, the same permissions and the same audit log as everyone else on your account. How authorization works

  1. 01

    No API key to copy

    You sign in with OAuth on Cademí itself and approve the connection. You never paste a key into a chat.

  2. 02

    Your permissions, nothing more

    The assistant acts as the administrator who approved it, on one account. The API checks every call.

  3. 03

    Every change on the record

    Each change lands in the account's audit log with the administrator who made it and what changed.

  4. 04

    It asks before it changes

    Every tool says whether it reads, writes or removes something, so your client can ask before a change and warn before a removal.

Connect in a minute

Add one URL to your assistant, sign in to Cademí and approve. Nothing to install and no key to paste.

  1. 1Open Settings, then Connectors.
  2. 2Choose Add custom connector and paste the server URL.
  3. 3Select Connect, choose your account and sign in as an administrator.
  4. 4Approve the connection. Then ask your first question.
Full setup guide →
Enterprise-grade audit

Every action has a name
and a trail behind it.

What your assistant does through Cademí MCP lands in the same audit log and request log as the API and the CLI. One trail, whichever way in, down to the field that changed.

  1. Question

    “Extend João’s access to Marketing Essentials until December 31.”

    in Claude
  2. Tool callset_student_enrollment_status
    Tool call ID3f9c2a1e…1c34Approved by you in the client
  3. API request

    PATCH/users/usr_1042/enrollments/enr_4821

    Tool call ID3f9c2a1e…1c34Request01J8Z3ZQ…YQF5Csucceeded · 200
  4. Audit entryaud_9183
    users.enrollments.update
    Actor
    adm_214
    Occurred at
    2026-10-14T13:05:12Z
    Resource
    enr_4821
    Outcome
    succeeded · 200
    Tool call ID3f9c2a1e…1c34
    What changed
    ends_at2026-09-122026-12-31

The same tool call ID travels from the call to the audit entry.

  • A real author for every action

    Each connection acts as one administrator on one account, through the OAuth authorization you approved. The audit log records that administrator as the author of every change, along with the client.

  • Permissions checked on every call

    The assistant has the administrator's permissions, re-evaluated on every tool call. Remove a permission in the dashboard and the next call no longer has it.

  • Every change, with its diff

    Each write, denial and background effect becomes an audit entry: actor, outcome, HTTP status, operation, resource, and what changed, before and after.

  • Reads are recorded too

    Every authenticated request goes to the request log, linked to the audit entries it produced.

  • Traceable end to end

    Each tool call has an ID that follows every API request it made, so one question to your assistant becomes a chain you can follow to each change.

  • Ask the trail itself

    With the audit.read permission, the assistant answers "who changed João's access yesterday?" by reading audit entries and requests, filtered by resource, action, outcome, request ID and period.

Questions

What is an MCP server?

The Model Context Protocol is an open standard that lets AI assistants use tools from other services. The Cademí MCP is a remote MCP server: your assistant connects to it over the internet and uses its tools to read and change your Cademí account.

Which account does the assistant work on?

The one you choose when you approve the connection, as the administrator who approved it. To work on another account, connect again and choose it.

Can the assistant delete data?

Only what your administrator permissions allow. Tools that remove or revoke something are marked as destructive, so your client can ask you first.

How do I remove access?

The connection acts as the administrator who approved it, so it stops working when that administrator is removed from the account, moved to another account, or has their access revoked.

How does the assistant know how to work with Cademí?

The server ships skills: step-by-step strategies to investigate an access problem, audit a student's experience and review a course, plus a guide to working safely with the server. Assistants that support MCP skills can load them, and the same strategies are available as prompts. Skills and documents come in English and Portuguese.

MCP, CLI or API?

Use the MCP to work with your account in a conversation. Use the CLI for scripts, CI and local event forwarding. Use the API to build the integration into your own systems. All three reach the same API v3.