Product Studio

Local Product Owner workspace: a browser chat UI, your company OpenAI-compatible LLM (no Cursor), and optional MCP tools for Jira, Confluence, GitLab, Workflow, Database, and Solver.

Requirements

• Node.js 22.13+

• Access to your LLM endpoint

• Optional: Jira / Confluence / GitLab — the app still runs if any are missing

Related MCP server: heimdall

Install

npm install
npm run build

Run (chat UI)

npm start
# → http://127.0.0.1:8787

Or:

npx tsx src/index.ts serve --port 8787

Login / members

Access requires a Studio account (httpOnly session cookie). Passwords are stored as scrypt hashes in SQLite — never plaintext.

First admin (pick one):

1. Open the UI with no users → Create admin setup screen

2. Or CLI:

npx tsx src/index.ts user bootstrap --username admin --password "your-secure-password"

Add a limited member (after admin exists):

npx tsx src/index.ts user create --username alice --password "alice-pass-here" --role limited
npx tsx src/index.ts user list

Or in the UI: Settings → Members (admin only).

Configure

Configs persist in SQLite (data/product-studio.sqlite).

In the UI: send /config and follow the menu.

CLI:

npx tsx src/index.ts config llm --base-url "https://your-llm/v1" --api-key "..." --model "..."
npx tsx src/index.ts config jira --base-url "..." --email "..." --api-token "..."
npx tsx src/index.ts config database --dialect sqlite --file-path data/app-data.sqlite --name default
npx tsx src/index.ts status

UI workspaces

• Chat — agent conversation (New chat keeps past threads; /start clears only the current chat; last 20 messages are sent as context)

• Workflows — create, edit, run, and visualize HTTP/SQL sequences; approve pending SQL here

• Settings — LLM / Jira / Confluence / GitLab / Database forms

• Status — connection state and active MCP tools

Workflows

npx tsx src/index.ts workflow create --name "demo" --description "sample" --definition-file data/sample-workflow.json
npx tsx src/index.ts workflow visualize demo

HTTP nodes support GET / POST / PUT / PATCH / DELETE with templated headers, query (object or query string), and body (JSON object or raw string). Body is not sent on GET/HEAD by default. Use {{nodeId.path}} and {{secrets.*}} in URL/headers/query/body. SQL nodes use type: "sql" (or "database") with arbitrary sql (and optional params / connectionId). Every SQL statement pauses for approval (including SELECT) until POST /api/approvals/:id/approve.

Return mapping (returns)

Optional top-level returns declares the Output shape of a successful run. Values are templates against node outputs ({{nodeId.path}}). A single template keeps the typed value; missing paths become null and are listed in meta.returnWarnings (set returnsStrict: true to fail instead).

When returns is present, MCP workflow_run / POST /api/workflows/:id/run include returns and omit full outputs unless you pass includeNodes: true (Studio Run always sets this for debugging). Workflows without returns keep the previous full-outputs response.

{
  "nodes": [ /* ... */ ],
  "returns": {
    "response": "{{get_profile.body}}",
    "token": "{{exchangeToken.body.access_token}}",
    "userId": "{{profile.body.id}}"
  }
}

HTTP node outputs always include body (parsed JSON or raw text), plus status, ok, text, and response headers. Use {{nodeId.body}} for the full payload object, {{nodeId}} for the entire node bag, or {{nodeId.body.field}} for a leaf. Body object fields are also promoted to the top level for older templates ({{nodeId.field}}), except when they collide with envelope keys (body, status, ok, text, headers).

Example GET (query) and POST (JSON body + headers):

{
  "nodes": [
    {
      "id": "listItems",
      "type": "http",
      "name": "List items",
      "method": "GET",
      "url": "https://httpbin.org/get",
      "headers": { "Accept": "application/json" },
      "query": { "page": "1", "q": "{{input.search}}" }
    },
    {
      "id": "createItem",
      "type": "http",
      "name": "Create item",
      "method": "POST",
      "url": "https://httpbin.org/post",
      "headers": {
        "Content-Type": "application/json",
        "Authorization": "Bearer {{secrets.jira.apiToken}}"
      },
      "body": { "title": "{{input.title}}", "from": "{{listItems.url}}" }
    }
  ]
}

Smoke-check HTTP request construction: npx tsx scripts/smoke-http-node.ts
Smoke-check workflow engine (transform, choice, parallel, auth): npm run smoke:workflow

Node cheat-sheet (extras)

Example SQL node:

{
  "id": "q1",
  "type": "sql",
  "name": "list users",
  "operation": "sql",
  "sql": "SELECT id, email FROM users LIMIT 20",
  "connectionId": "default"
}

Approvals API: GET /api/approvals?status=pending, POST /api/approvals/:id/approve, POST /api/approvals/:id/reject.

MCP servers

Database MCP tools (English names): database_list_connections, database_upsert_connection, database_delete_connection, database_test_connection, database_list_databases, database_list_schemas, database_list_tables, database_describe_table, database_list_indexes, database_list_foreign_keys, database_request_sql (queues approval), database_list_approvals.