Work with the MCP server

Has your coding agent read Mindset's current method before it builds, start from worked examples, and hand a person's decisions back to them.

The mindset-admin server carries Mindset's own method for every job, and it always matches the platform version you're connected to. Your coding agent builds the right thing first time when it reads that method before building, instead of relying on what it remembers about Mindset.

Ask your coding agent#

Before you build anything in Mindset, read get_information's topic index and the topics for the job. Add that rule to CLAUDE.md so you follow it in every session.

Skill: mindset-mcp

Read the method from the server#

get_information with no arguments lists every topic and says when to read it. These matter most:

Topic	When your coding agent reads it
`what-is-this-platform`	Once, before its first build
`choosing-what-to-build`	Before any kind of build it hasn't done before. See Choose what to build
`authoring:<type>`	Before creating that type, such as `authoring:script` or `authoring:connection`
`boundaries`	When a request might need something Mindset can't do, so it says no plainly instead of building around it
`behaviour-testing`	Before writing or running behavior tests
`environments`	Before naming an environment in a call

Start from an example#

find_examples returns worked builds close to what you describe. Searching with no resource_type first lets your coding agent find out what to build before it knows. Names in an example come from the build it was drawn from, so your coding agent reads the real operation with get_resource before copying a call. An example marked counter-example is a refusal to pass on to you, never a payload to copy.

One environment per call#

The server address names the environment every call uses. An org admin's signed-in connection can point a single call at another environment with envSlug, and nothing is remembered between calls. A session using an API key can only act in its own environment.

A person's decisions stay with a person#

Your coding agent asks and waits for each of these:

Creating the org, an environment or an org API key. These are done in the Studio, and no tool creates a key.
Entering a credential. Your coding agent calls request_capture and gives you a capture URL. You enter the value there, and it never comes back to the conversation.
Approving a write operation, when your org doesn't turn new ones on automatically.
Activating a version of an agent people already use.
Opening an agent to The Hub, Embedded in your own product or Outside systems.

Report what doesn't work#

When something on the platform fails and your coding agent can't make it work, it files platform_issue with what it tried and what failed, even after finding a workaround.

Done when#

Your project's CLAUDE.md tells your coding agent to read get_information before authoring.
Your coding agent names the topics it read when it proposes a build.