Connected AI · MCP reference

21 tools your AI agent can use. Zero rules it can break.

Connect Claude, ChatGPT, Codex, or Cursor and it works on the same experiment record as your team: 7 tools to read it, 12 to update it, and 2 to delete from it if you grant that scope at connection. Every write follows the same rules the app enforces: no skipped stages, no experiment closed without a recorded result and learning. This page is the exact boundary.

Why the rules matter

An agent with access to your tracker will happily mark a test as won. Not out of malice, out of confidence. Through MCP, Hypotize refuses that write: an experiment cannot skip a stage, and it cannot close without a measured result and a written learning. Your agent gets hands, not authority.

For anyone who wants to check the details

The 21 connector tools

Each tool has one defined job. Seven read the record, twelve update it, and two delete from it, off by default until you grant that scope at connection. Every write passes the same validation as the app itself: an experiment cannot skip a stage, and it cannot be closed without a result and a learning. That applies to your AI exactly as it applies to your team.

Look things up

hypotize_projects_list

List the projects you can access.

hypotize_project_get

Load one project's context, writing language, and prioritization framework.

hypotize_experiments_list

Filter a project's experiment backlog and pipeline.

hypotize_experiment_get

Read an experiment, its framework score, action plan, latest 50 comments, and writing language.

hypotize_learnings_list

Read outcomes, evidence, learnings, and next steps.

hypotize_tags_list

List the reusable tags of a project.

hypotize_frameworks_list

List a project's custom scoring frameworks.

Update the record

hypotize_project_create

Create a new project. Solo and Team have no project limit.

hypotize_project_update

Update project context and the language your agent should write in.

hypotize_experiments_import

Import up to 50 existing experiments, variants included, in one all-or-nothing batch.

hypotize_idea_create

Create a new idea in the project backlog, with its variants if it is an A/B test.

hypotize_experiment_update

Update an experiment without silently overwriting newer work.

hypotize_vote_upsert

Record your vote in the experiment's framework and recalculate priority.

hypotize_experiment_transition

Move an experiment through a validated lifecycle transition.

hypotize_experiment_record_outcome

Save the result, winning variant, learning, evidence, and next steps together.

hypotize_comment_add

Add a dated comment, such as an interim reading while a test is running.

hypotize_experiment_action_plan_set

Set the experiment action plan and estimated hours.

hypotize_tag_create

Create a reusable tag, or return it if it already exists.

hypotize_framework_create

Create a custom scoring framework with its own dimensions and formula.

Delete records

Opt-in. Off until you grant the delete scope at connection.

hypotize_experiment_delete

Permanently delete an experiment. Only with the delete scope you grant at connection.

hypotize_project_delete

Permanently delete a project. Only with the delete scope you grant at connection.

Updates are protected against duplicate requests and against silently overwriting newer work.

Connect your agent

A few steps, no API key

Pick your agent and follow its steps. Hypotize signs it in with OAuth: you choose what it may do, and you can revoke it from your project's Connections page at any time.

Claude

claude.ai, Claude Desktop and Cowork

1
Open Claude with Hypotize filled in
The button opens the Add custom connector dialog with the name and URL already set. By hand: Settings › Connectors › Add custom connector, then paste the URL below.
Add Hypotize in Claude
```
https://api.hypotize.ai/mcp
```
Click Add, then Connect
Claude adds the connector and asks you to connect it. Connectors added in claude.ai also work in Claude Desktop and Cowork.
Authorize in your browser
Sign in to Hypotize and choose what the agent may do. Delete access stays off unless you tick it. The client then shows up in Connections, where you can revoke it at any time.

Outside the connector

What a connected client cannot do

Connecting an agent does not grant direct access to the Hypotize database or every feature in the app. These operations are not exposed through MCP.

Administer people or the account

The agent cannot invite, remove, or change the role of anyone, and it cannot touch billing. People and account administration happen in Hypotize, by a person. The agent can see who owns what, so it can answer questions about assignments.

Rename or merge tags across records

Tags and custom frameworks can be listed and created, and tag values can be written when importing experiments, creating an idea, or updating an experiment. What no tool does is rename or merge a tag across every record that uses it: that stays a deliberate action in the app.

Delete an account

No MCP tool deletes an account. Deleting a project or an experiment needs the delete scope, off by default until you grant it at connection, and even then it respects your role. A client with update permission can also move an experiment to Archived, which keeps it in the project record without deleting it.

Access files or edit existing comments

The connector cannot upload files, browse storage, or perform unrestricted file operations. It can record evidence URLs, paths, or external IDs. Experiment details include the latest 50 comments and the total count. The agent can add dated comments, but cannot edit or delete them.

Related evidence

See the record the tools read and update

The experiment memory guide documents the record fields and lifecycle. The tracker comparison explains when this connector is useful and when a spreadsheet or document is enough.

Experiment memory guide Compare tracker options