The Digital Factory Senegal team is extremely proud to share with you today our open-source contribution to the world of AI and ERP: The very first MCP server for Dolibarr!
What is it?
For those who use the Model Context Protocol (MCP) ecosystem, we have developed a public Node.js server that directly connects your favorite AI assistants (Claude Desktop, Cursor, Windsurf) to your Dolibarr instance via its native REST API.
By installing it, your AI instantly becomes capable of:
Creating and validating invoices / drafts
Consulting your bank accounts and editing expert summaries of the General Ledger
Managing your CRM (Third parties, Contacts) and your Commercial Proposals
Tracking your stock inventories and querying the system on any accounting parameter.
(The project integrates more than 55 direct action tools).
How to use it (100% Free & Open-Source)?
No PHP module to install! Just activate the REST API of your Dolibarr, get your key, and execute the universal command:
Great contribution, @falldaouda! This is exactly the kind of tooling the Dolibarr ecosystem needs.
A few thoughts from deploying MCP + Dolibarr in production with clients:
What works really well
The read-side tools (fetching invoices, contacts, products via REST) are solid for agentic workflows. Once Claude has the right Dolibarr context it navigates the data model surprisingly well.
Where RAG adds extra leverage
For clients with large product catalogs or complex pricing rules, pairing MCP with a RAG layer (embedding Dolibarr data + internal docs) dramatically improves accuracy. The MCP handles live CRUD while RAG handles “which product fits this client?” style lookups — the combination is powerful.
On the write side
The trickiest part with write operations (create invoice, update stock) is keeping the AI conservative enough — hallucinated line items in production are painful. Are you using any confirmation/validation step before writes execute in your server?
I’m currently building similar integrations for several Dolibarr clients and would love to contribute or collaborate. What’s the best way to get involved?
Really interesting to see this gain traction — thanks for sharing it, @falldaouda.
One thing worth flagging for anyone integrating AI with Dolibarr: the MCP / tool-calling approach (where the AI calls REST endpoints directly) is excellent for structured, action-based tasks — creating invoices, querying CRM, updating records. Where it gets interesting is pairing it with a retrieval layer. Dolibarr’s REST API doesn’t expose unstructured business context — historical patterns, cross-entity analytics, the kind of data that’s useful for natural-language questions over your own ERP data. I’ve been working on RAG-based approaches that sit alongside the REST API layer, so the AI can both act (via MCP tools) and reason over your business data (via retrieval). The two layers handle quite different question types.
For anyone setting this up: make sure your Dolibarr API key has the correct permission scope (read-only vs full access), and keep it out of shared config files. Also worth checking server-side rate limits if the AI starts making rapid bulk calls.
Excited to see where this goes — the Dolibarr + AI ecosystem is still early but moving fast.
