odoo-mcp
Gives an LLM agent read and write access to partners, sales orders, stock levels, and invoices in an Odoo instance.
README
odoo-mcp
An MCP server for Odoo. It gives an LLM agent read and write access to partners, sales orders, stock levels and invoices in an Odoo instance.
All six tools work against the Odoo in docker compose: answers are capped, every call
has a deadline, ODOO_READONLY=true leaves the write tools unpublished, and the server
talks either over stdio or over http. What is missing is listed at the bottom.
Why
Odoo is a full ERP with a wide API, which makes it hard for an agent to use. Its
external API is XML-RPC with generic search_read and execute_kw calls, so the agent
would need to know model names, field names and domain syntax. This server exposes a
small number of named tools instead, each one covering a task a person would actually
ask for.
Odoo Community runs in Docker with demo data, so the whole thing can be tried without any real system.
Tools
| Tool | What it does |
|---|---|
search_partners |
find customers and suppliers |
get_sales_order |
read an order with its lines and status |
create_sales_order |
create a quotation for a customer |
confirm_sales_order |
turn a quotation into a confirmed order |
get_stock |
current quantity on hand for a product |
list_invoices |
invoices for a partner, filtered by state |
Quick start
make up # Odoo Community with demo data, waits until it answers
make seed # one customer, one product, a known quantity of it
make demo # the whole path, printed as it happens
The first make up takes a couple of minutes: Odoo builds its database and loads the
demo data. Everything after that is seconds.
To run the server itself rather than the demo:
make run # over stdio, the way a desktop client starts it
docker compose up -d mcp # over http, then curl localhost:8080/health
Port 8080 is a busy one. If something else on the machine already holds it, set
MCP_HOST_PORT to anything free and compose will use that on the host side.
Demo
make demo starts the server as its own process, talks to it over stdio exactly as an
MCP client does, and walks the path a salesperson walks. Every line below came back over
the protocol:
Connected. Tools on offer: confirm_sales_order, create_sales_order, get_sales_order, get_stock, list_invoices, search_partners
1. Who is MCP Demo Customer?
MCP Demo Customer, Vienna, id 44
2. What is on the shelf for MCP-DESK?
[MCP-DESK] MCP Demo Desk: 42 on hand, 42 free, in YourCompany
3. Order 2 of them.
S00095 for MCP Demo Customer: quotation, 1147.7 USD
4. Confirm it.
S00095 is now confirmed
5. Read it back, lines and all.
2 x [MCP-DESK] MCP Demo Desk = 998
6. And what a refusal sounds like.
Error executing tool confirm_sales_order: Order S00095 is confirmed. Only a quotation can be confirmed.
The order number climbs with every run, the rest stays put: that is what make seed is
for, and running it twice changes nothing.
How it works
The server connects to Odoo over XML-RPC. Each tool builds the domain and the field list itself, so the agent never sees Odoo internals. Results are trimmed to the fields that matter and returned as flat records.
Design decisions
Named tools, not a generic query tool. A single run_query tool would be easier to
write and much worse to use. Named tools carry their own validation and make it obvious
what the agent is allowed to do.
Read and write are separated. The two write tools are marked destructive in their MCP annotations and the four read tools are marked read-only, so a client can ask its user before an order is created or confirmed.
Arguments are checked before Odoo is called. Odoo takes an order line of zero quantity without a word, and answers an unknown id with its own record ids in the text. Both are caught here, so a refusal says what to do instead of what broke inside the ERP.
A cut answer says it was cut. No call hands back more than 200 records. When more
matched, the answer carries truncated: true and a hint on how to narrow the request,
because twenty results out of thirty-eight look exactly like all of them otherwise.
Read-only mode removes the write tools rather than blocking them. With
ODOO_READONLY=true they are never published, so a client sees four tools and a call by
name comes back as an unknown tool.
Errors are translated. Odoo raises faults with long tracebacks. The server extracts the message and returns it in a form the agent can act on.
Health says nothing about Odoo. /health answers for the process alone. The compose
healthcheck watches it, and an Odoo that is briefly away must not get a working server
restarted underneath it. Whether Odoo answers is visible on the first tool call.
What is not here yet
- paging through a cut answer: a tool says there is more, but there is no way to ask for the next page yet, only a narrower question
- multi-company support: a company is pinned with
ODOO_COMPANY_IDSand every answer names its own, which is not the same as running across several of them - cancelling an order, and editing one that already exists:
action_cancelover XML-RPC returns a dialog to open and leaves the order as it was - creating a partner, so an agent can sell to someone who is not in the database yet
- authentication on the http transport: it is meant for a private network, not the open one
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
E2B
Using MCP to run code via e2b.