<!-- generated by scripts/gen-reference.mjs; do not edit -->

MCP tools and prompts

Tools registered by the Basics MCP server. Call them over stdio (basics-mcp) or HTTP (basics ui).

get_vault_overview

Read the whole vault catalog in one call: every layer with its techs (kind, use_when, avoid_when, performs_best), the stack profiles and the past projects. Use it to reason over the vault when a ranked shortlist is not enough; read deciding notes with get_note before choosing. Read-only.

Input	Type	Required	Description
layers	string[]	no	Restrict the catalog to these layer ids

Minimal call:

{
  "name": "get_vault_overview",
  "arguments": {}
}

list_layers

List the stack layers (backend, frontend, testing, ...) and the technologies stored under each.

No inputs.

Minimal call:

{
  "name": "list_layers",
  "arguments": {}
}

search_vault

Full-text search over the user's stack vault. Use before proposing new notes to avoid duplicates.

Input	Type	Required	Description
query	string	yes	—
type	layer | tech | practice | check | profile | project	no	—
layer	string	no	Restrict to tech notes in this layer
limit	integer	no	—

Minimal call:

{
  "name": "search_vault",
  "arguments": {
    "query": "rest api"
  }
}

get_note

Read a vault note (frontmatter + markdown body) plus which notes link to it.

Input	Type	Required	Description
id	string	yes	—

Minimal call:

{
  "name": "get_note",
  "arguments": {
    "id": "fastapi"
  }
}

recommend_stack

Given a natural-language project description, return ranked technology candidates per layer and matching stack profiles from the user's vault. Treat as a shortlist: read the candidate notes, choose, and explain the choice to the user. Prefer the user's vault over general knowledge.

Input	Type	Required	Description
description	string	yes	—
layers	string[]	no	—
per_layer	integer	no	—

Minimal call:

{
  "name": "recommend_stack",
  "arguments": {
    "description": "example"
  }
}

get_stack_context

Compile best practices and validation checks for a chosen set of technologies into one markdown document. Call once the stack is chosen and follow it while writing code; run the listed checks before declaring work done.

Input	Type	Required	Description
tech_ids	string[]	yes	—
versions	map<string, string>	no	—

Minimal call:

{
  "name": "get_stack_context",
  "arguments": {
    "tech_ids": [
      "example"
    ]
  }
}

get_check_templates

Return the file templates (tests, configs) shipped by the checks of a chosen stack. Write missing files into the project; never overwrite existing files.

Input	Type	Required	Description
stack	string[]	yes	—
versions	map<string, string>	no	—

Minimal call:

{
  "name": "get_check_templates",
  "arguments": {
    "stack": [
      "example"
    ]
  }
}

detect_stack

Detect technologies and versions from a project directory's package.json, pyproject.toml and requirements files. Read-only.

Input	Type	Required	Description
path	string	yes	Existing project directory

Minimal call:

{
  "name": "detect_stack",
  "arguments": {
    "path": "./project"
  }
}

propose_note

Propose a new note or an update to an existing one. It lands in the vault inbox for the user to approve; it is NOT applied directly. Frontmatter must match the note type schema (e.g. tech: kind, layer, use_when, avoid_when, performs_best, practices, checks). Use [[id]] wikilinks in the body to connect notes.

Input	Type	Required	Description
id	string	yes	Slug, e.g. 'fastapi' or 'react-server-components'
type	layer | tech | practice | check | profile | project	yes	—
frontmatter	map<string, unknown>	yes	Fields for the note type, excluding 'type'
body	string	yes	Markdown body
reason	string	yes	Why this note should be added/changed
agent	string	yes	Name of the proposing agent, e.g. 'claude-code'

Minimal call:

{
  "name": "propose_note",
  "arguments": {
    "id": "fastapi",
    "type": "layer",
    "frontmatter": {},
    "body": "example",
    "reason": "example",
    "agent": "claude-code"
  }
}

record_project_outcome

Propose an outcome for a project note: append lessons, optionally extend tech notes (use_when, avoid_when, performs_best, lessons) and add a dated outcome section. Everything lands in the vault inbox for review; nothing is applied directly.

Input	Type	Required	Description
project	string	yes	Project note id
summary	string	yes	What happened and what was learned
lessons	string[]	no	—
tech_updates	{ id, add_use_when, add_avoid_when, add_performs_best, lesson }[]	no	—

Minimal call:

{
  "name": "record_project_outcome",
  "arguments": {
    "project": "fastapi",
    "summary": "example"
  }
}

export_project

Render the files that make a project's coding agents follow the user's stack: STACK.md always, plus AGENTS.md, CLAUDE.md and/or Cursor rules. Pass the current file contents in existing so agent-file blocks merge instead of overwriting user text; basics never writes outside the vault.

Input	Type	Required	Description
stack	string[]	yes	—
targets	agents | claude | cursor[]	no	—
existing	map<string, string>	no	Current file contents keyed by relative path

Minimal call:

{
  "name": "export_project",
  "arguments": {
    "stack": [
      "example"
    ]
  }
}

list_proposals

List agent proposals waiting for user review in the vault inbox.

No inputs.

Minimal call:

{
  "name": "list_proposals",
  "arguments": {}
}

MCP prompts

Prompts render a ready-to-use instruction; the client inserts it into the conversation. Discover them with prompts/list and render one with prompts/get.

plan-stack

Plan a project's stack from the vault: read the whole catalog and the ranked shortlist, decide one tech per layer, explain the choice, propose any note the vault is missing for that decision and ask for confirmation before writing files.

Argument	Required	Description
project	yes	What the user is building: product, users and any constraint already known.