Coteach MCP server icon

Coteach

by Coteach

HIPAA CompliantSOC2 ReadyISO 27001 Ready
Data & Research1 tool

Generate classroom-ready K-12 math diagrams from an AI agent. One tool, one operation, no read path into Coteach. The Coteach platform scans and scores student work; the connector reaches none of it. OAuth sign-in, identity-only scopes.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

Connect Coteach via MCP

https://coteach.ai/api/mcp

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

Coteach Tools & Capabilities (1)

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • Coteach publishes no tool list. Its connector documentation, blog and llms.txt describe the capability in prose and never name a tool, so the one name comes from Anthropic's directory. There is no second enumeration to diff against. What Coteach does publish is a partial negative enumeration — an explicit list of what the connector does not access — which is the strongest boundary evidence here but covers the agent side, not the Coteach side.
  • We could not read the tool's schema or annotations. The endpoint returned 401 to an anonymous request, so no parameter-level detail is published here and the classification is ours from the name.
  • Anthropic's directory records the permission as "Read and write". Coteach's documentation describes neither a read of your account nor a write into it — the tool returns images to the conversation. The two do not agree, and the directory's label is the broader of the two. We could not resolve which is right without a schema.
  • No scope separates anything. The resource descriptor advertises only openid, email and profile; the authorization server adds offline_access. All are identity claims. With one generate tool the practical consequence is small, but nothing at consent constrains the connector.
  • It is generate-only, which is the point and also the ceiling. You cannot search diagrams you made before, pull a saved worksheet into the conversation, or save the output back into Coteach. Every session starts blind to what Coteach already holds, and the images live in the conversation until you download them.
  • Coteach's own llms.txt carries a stale reassurance. Its "No student data: Does not collect or process student information" line is contradicted at the platform level by Coteach's product post and privacy policy. It is accurate only as a description of the diagram capability.
  • The scope is narrow today by Coteach's own account, and Coteach says it intends to widen it. Its launch post describes connecting exit tickets, assessments and student work as future work. Treat "no student data" as a finding about 2026-08-23, not a structural property.
  • Coteach publishes no independent privacy certification. It asserts FERPA, COPPA and SOPIPA compliance in its privacy policy, but names no external attestation such as SOC 2, iKeepSafe or 1EdTech. Its subprocessor list is available only on request to privacy@coteach.ai.
  • The daily cap is invisible to the agent. 100 diagrams per day at launch, with no tool that can read remaining quota, so an unattended agent generating diagram sets in a loop will hit it blind.
  • Accuracy is high but not total, and the failure is quiet. Coteach reports an 88.8% pass rate on its own benchmark, which it states is weighted toward hard cases. A wrong diagram is mathematically wrong rather than obviously broken — Coteach's own post shows a competitor's figure labelling two visibly different lengths as equal. Check a diagram before putting it in front of a class.
  • Duplicate generation is not addressed. Coteach's documentation says nothing about idempotency or about whether a retried request after a timeout consumes a second slot of the daily 100. Assume it can.
  • Retention of generated images is vague. Coteach states generated diagram assets "may be temporarily stored" so they can be previewed and downloaded, and does not define how long.

Frequently asked questions

No. The one tool Anthropic lists generates math diagrams from a text description, and Coteach's connector documentation describes only that. Coteach states the connector does not access Claude memory, chat history, uploaded files, or unrelated user data. It names no read path into a Coteach account at all, so there is no route by which a roster or a scored assessment travels into the agent.

Yes, and the two facts must be held together. Coteach's July 2026 product post describes teachers scanning completed paper Cool Downs so Coteach reads each response, suggests a score and tracks each student's progress over time. Coteach's privacy policy defines Student Data and commits to handling it under FERPA. The connector is scoped away from all of that.

Anthropic's directory lists exactly one, named CreateDiagrams. Coteach's documentation describes a single diagram generation capability and never names a tool, so there is no vendor list to check the name against. The tool names a specific object and one action on it, and takes no service, action or endpoint selector, so it is not a dispatcher onto a wider API.

Coteach states it complies with both. Its privacy policy says it acts as a school official under FERPA when processing Student Data under School agreements, and names COPPA and state laws such as SOPIPA. Coteach publishes no independent certification or third-party audit for those claims, which is a weaker position than a vendor holding an external attestation.

Only identity claims. Coteach's RFC 9728 resource descriptor declared openid, email and profile on 2026-08-23, and its authorization server adds offline_access. None of those names an application operation. With a single generate tool the practical consequence is small, and no scope names a verb the tool does not carry.

Yes. Coteach's connector documentation states that at launch, connected users can create up to 100 diagrams per day under fair-use limits. No tool in the listing can read a remaining quota, so an agent cannot check how many diagrams are left before generating more. Coteach publishes no charge to the teacher for a connector diagram.

Nothing in the listing writes to your account. The single tool returns generated diagram images into the Claude conversation rather than saving them into Coteach. Coteach states generated diagram assets may be temporarily stored so they can be previewed and downloaded. Anthropic's directory records the permission as Read and write, which is broader than the documented behaviour.

A Coteach account and a Claude account. Coteach's documentation says to open Settings then Connectors in Claude, search for Coteach, select it and connect your Coteach account. Coteach is a launch connector for Claude for Teachers, which Anthropic offers free to verified K-12 educators in the US, but the connector itself does not require that programme.

Sources

Use in Agentman

Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.

Open in Agentman Studio

Resources

Server Info

Category
Data & Research
Developer
Coteach
Tools
1
Domain
coteach.ai

Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.

Similar MCP Servers for Data & Research

Box logo

Box

by Box

Coming Soon
Data & Research10 tools
View detailsConnect
Candid logo

Candid

by Candid

Data & Research6 tools
View detailsConnect
GoDaddy logo

GoDaddy

by GoDaddy

Data & Research2 tools
View detailsConnect

Related Resources

Browse All MCP ServersClaude & ChatGPT Setup GuideBrowse AI Skills