How Mindset fits together

Mindset holds your agents and the plumbing they run on. Builders and operators change it through the MCP server, and your product reaches agents through the SDK and API.

Mindset holds every agent's configuration and runs the plumbing a production agent needs: connections and their credentials, versions, testing, run records, cost and audit. Your team builds none of it, and spends its time on what the agents do.

Mindset#

Page	Covers
Agents	A prompt, a model and, optionally, a script, using only the resources you give it
Scripts	An order of phases an agent must follow, each with a gate
Connections and operations	Links to outside systems, their credentials, and the reads and writes agents make through them
Functions	Fixed steps with one right answer, which agents call as tools
Widgets	Tables, cards, forms and charts an agent shows in a conversation
Versions, testing and going live	Every save is a version, and an untested operation can't go live
Runs	Each execution of an agent, recorded from what started it to how it ended
How agents are started and reached	Which ways in an agent is open to
Environments	Sandbox and Production, kept apart in one org
Sessions and trust	How a user of your app talks to an agent while your API key stays on your server
Security and approval	The rules Mindset enforces for every agent, and how writes are approved

The MCP server#

Builders and operators work on Mindset from a coding agent or AI client, through the mindset-admin MCP server.

Builders ask a coding agent such as Claude Code or Cursor to create agents, connect systems and write tests. The Mindset skills tell it how to do each job. See Start with your coding agent.
Operators ask what an agent did, why a run failed or what it costs, then test and roll out the fix. They can do the same in the Studio. See Check what happened.

Members use the mindset MCP server, which lists the agents available to them and hands one a task. See Use your agents from AI clients. MCP tools lists every tool on both servers.

Why Mindset is MCP first#

A coding agent such as Claude Code or Cursor already reads your code and changes it. Give it the Mindset skills and the MCP server, and it also configures the agents your code calls, in the same session.

Without Mindset, the coding agent writes the plumbing into your codebase, and your team maintains it from then on. With Mindset, the plumbing is already running and the coding agent configures it.

Without Mindset	With Mindset
Credential handling written into your code	A person enters each key on a capture page, and it never reaches the coding agent
Versioning and rollback built per project	Every change is a version, activated or rolled back with no release of your code
Run logs and cost tracking built and hosted by you	Every run is recorded, with each model call, tool call and its cost
Rules copied into each codebase	One set of rules Mindset enforces for every agent, and no agent can approve a write operation

The MCP server gives the coding agent what it needs to work safely: Mindset's current method for each job, the agent's versions, and which environment it's working in. Your rules stay in one place while coding agents across your team build on them, and your app reaches the result through the SDK and API.

The SDK and API#

Your product reaches agents through the SDK and API.

The SDK puts an agent on your page. It runs the agent in the browser, so replies stream with low latency and page tools run with no round trip. Your server makes one call to create a session with your org API key, which never reaches the browser. See Put an agent on your page.
The API starts an agent from your server with an org API key, with no person in a chat. Schedules and other products' webhooks start runs the same way. See Start it from your server or a schedule.

People in your org also chat with agents in the Hub. Every way in reads the same configuration from Mindset, so a change to an agent needs no release of your code.