There is a specific point in an Odoo development session where the agent becomes genuinely useful instead of just fast: when it stops guessing about your data and starts reading it directly. You ask “which sales orders are overdue this month?” and instead of giving you a template query, it runs the query and hands you the results.
That point requires two things: Claude Cowork as your agent, and a configured odooly skill pointing at your instance. This post walks through both.
1. What Is Claude Cowork
Claude Cowork is Anthropic’s desktop application for agentic knowledge work. It is the evolution of Claude Code — bringing the same autonomous, multi-step execution modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or… out of the terminal and into a desktop environment built for anyone who works with files, documents, and data every day.
Unlike the web chat interface, where every step requires a promptThe input text provided to an LLM to guide its response. Prompt design — choosing words, structure, and examples — significantly affects output quality. Also referred to as the user message or query. and a reply, Cowork is designed for delegation: you describe the outcome you want, and Claude works through the task independently, reading and writing local files, running tools, and delivering a result. You do not coordinate each step. You come back to find it done.
For Odoo teams, this distinction matters. Asking a web chat interface “show me overdue invoices” produces a query you have to run yourself. Asking Cowork the same question — with the odooly skill installed — executes the query against your live instance and returns the data directly.
Cowork also has a close sibling: Claude Code, Anthropic’s terminal-based CLI for developers. Both share the same skills infrastructure. A skill installed in Claude Code is available across every project and session on that machine; the same skill installed in Cowork is available inside the desktop app. The installation steps in this post work for both.
2. Installing the Trobz Public Skills Plugin
The odooly skill lives in the Trobz public skills marketplace, a repository of reusable agent skills for Odoo development and document processing. In Claude Cowork, you install it through the GUI in five steps.
Step 1 — Open Customize
Click Customize in the left sidebar of Claude Cowork.
Step 2 — Add a marketplace
Click the + button next to Personal plugins, then choose Create plugin → Add marketplace.
Step 3 — Add from a repository
In the Add marketplace dialog, select Add from a repository.
Step 4 — Enter the Trobz repository URL
Paste https://github.com/trobz/public-skills into the URL field and click Sync.
Step 5 — Install the Odoo plugin
The Odoo plugin by Trobz appears in the directory. Click + to install it.
The plugin includes several skills. The one we need here is odoo:odooly.
3. What the Odooly Skill Does
Once installed, the odoo:odooly skill lets Claude Code query and inspect live Odoo data using the odooly CLI. You ask for data in plain language, the agent constructs the right odooly command, runs it, and returns the results.
A few examples of what this looks like in practice:
- “List the 10 most recent sale orders in draft state”
- “Find all partners from Vietnam with a credit limit above 1000”
- “Show me the installed modules on staging”
- “Check the
payment_term_idfield on invoice number INV/2026/0042”
The skill maps natural language references to Odoo modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or… names automatically (partners → res.partner, sales orders → sale.order, invoices → account.move). It always reads from ~/odooly.ini for connection details, which means credentials never appear in the conversation or in a committed file.
4. Creating the ~/odooly.ini Configuration File
odooly reads connection details from ~/odooly.ini — one section per Odoo instance or environment. This file lives outside any project directory and should never be committed to git.
[staging]
scheme = https
host = HTTP_AUTH_USER:[email protected]
port = 443
username = admin
password = ADMIN_PASSWORD
database = project_staging
protocol = jsonrpc
[production]
scheme = https
host = HTTP_AUTH_USER:[email protected]
port = 443
username = admin
password = ADMIN_PASSWORD
database = project_production
protocol = jsonrpc
Each section name (staging, production) becomes the environment identifier used in queries. The parameters:
| Parameter | Description | Example |
|---|---|---|
scheme |
Connection protocol | https or http |
host |
Server hostname, may include HTTP auth | user:[email protected] |
port |
Server port | 443 (HTTPS), 8069 (local) |
username |
Odoo username | admin |
password |
Odoo user password or API key | your_password |
database |
Database name | company_production |
protocol |
Odoo RPC protocol | jsonrpc |
For Odoo.sh or instances behind HTTP authentication, include the credentials directly in the host field. Keep ~/odooly.ini secure — it contains plain-text credentials.
5. Connecting and Querying from Claude Cowork
With the plugin installed and your ~/odooly.ini ready, set up a Cowork project that includes the config file, then start asking questions.
Step 1 — Create a new project
Click Projects in the left sidebar, then click New project in the top right.
Step 2 — Configure the project and attach odooly.ini
Name the project (e.g. “Odoo”), then add your odooly.ini file under Add files. Cowork will make it available to all tasks in this project — no need to upload it each time.
Step 3 — Start a task and select your project
Click New task, then click Work in a project at the bottom of the chat input and select your Odoo project. Cowork loads the project’s files and instructions for the session.
Step 4 — Ask your question in plain language
No slash command needed. Describe what you want and Cowork figures out the rest:
Do we know about BENK company?
The agent runs the appropriate odooly commands and returns a structured result — contact details, credit limit, type, and location — without you writing a single domain filter.
What This Changes
The practical shift is from describing your data to reading it. Before this setup, debugging an issue with a specific invoice or tracing why a workflow triggered incorrectly meant pasting records into the chat manually — or describing them from memory. With the Trobz odooly skill configured, Claude Cowork reaches the data directly.
That is the point where Cowork stops being a faster way to write queries and starts behaving like a second developer who actually knows your instance — its data, its state, its installed modules.
The full skills marketplace is at github.com/trobz/public-skills. The odooly documentation is at odooly.readthedocs.io.