Skip to content
MAIA LogoDocs

MCP

MCP prompts

MAIA's MCP connector ships five prompts, which are ready-made requests that tell your client which tools to call and in what order. In Claude Code they appear as slash commands.

What a prompt is#

A tool does one thing. A prompt is a short script for your client: it names the goal, the tools to use, and the point at which to stop and ask you. You fill in the arguments and the client does the rest.

Every prompt tells your client to bring MAIA's questions to you, and to show you a skill or a note before it saves one. How closely a client follows that is up to the client, so read what it shows you before you say yes.

Run a prompt#

In Claude Code, type / and pick the prompt from the list. It appears with the connector's name in front of it. Give the arguments after the command, in the order the tables below list them. Not every client shows prompts. The tools work without them.

The five prompts#

build_datasetWhat it doesStarts a new project and sees it through to a finished datasetArgumentscounty, request
review_projectWhat it doesChecks what an existing project contains and what may be missingArgumentsproject
run_skillWhat it doesRuns one of your skills in a county, as a new projectArgumentsskill, county
create_skillWhat it doesTurns a repeatable way of working into a skillArgumentsworkflow
teach_maiaWhat it doesGives MAIA a short profile of your workArgumentsNone

build_dataset#

countyMeaningThe county to work inExampleBoulder, CO
requestMeaningWhat to findExamplevacant parcels over 5 acres
  1. The client finds the county with find_county.
  2. It starts the project with create_project and follows the run with get_run.
  3. It brings every question and approval to you before it calls respond. Approving research above your row limit spends credits.
  4. When the run ends, it summarizes what MAIA built and gives you the project link.

review_project#

projectMeaningThe project's name

The client finds the project with list_projects, describes its layers and views with get_project, and reports what the dataset covers, which values look wrong or missing, and what to add next. It starts no run. To check values and their reasoning yourself, open the project link. See Review.

run_skill#

skillMeaningThe skill's nameExampleIndustrial infill filter
countyMeaningThe county to run it inExampleBoulder, CO

The client checks the name with list_skills, finds the county, and calls create_project with the skill. It runs your skills and your workspace's, not MAIA's ready-made ones. This is how a method you wrote for one county reaches the next. A threshold on a column the new county does not hold returns nothing, so ask MAIA to profile the county first when in doubt. See Compound.

create_skill#

workflowMeaningWhat the skill should do. Write "this conversation" to capture the work done in the current one

The client calls list_skills first, so it improves a skill you already have instead of making a copy. It drafts the skill the way MAIA writes one: no county, city, or project named, and every threshold spelled out. It shows you the name, the description, and the instructions, and it calls save_skill only after you agree. Saving with an existing name replaces that skill. See Skills.

teach_maia#

No arguments. The client writes a short work profile from what it knows about you: your role and company, the markets and counties you work in, the property types you look for, your criteria with their numbers, and the data you usually need.

It leaves out anything personal that is not about the work, and it asks about gaps instead of guessing. It shows you the profile, changes it until you approve, and then calls save_to_maia_memory with the title "Work profile". Running it again replaces that section.

Limits#

  • Skills saved this way are personal. Workspace skills are edited in MAIA.
  • Not every client shows prompts. The tools work without them.

Related#

  • Tools documents each tool a prompt calls.
  • Skills and Knowledge explain what create_skill and teach_maia write.