Dolibarr MCP Server

Dolibarr MCP delivers a Model Context Protocol (MCP) interface for the Dolibarr ERP/CRM. The project mirrors the project structure of prestashop-mcp: an async API client, a production-ready STDIO server, and focused documentation.

Design Note: While sharing the same architecture, this server implements specialized search tools (e.g., search_products_by_ref, resolve_product_ref) instead of a single unified get_ tool. This design choice ensures efficient server-side filtering via Dolibarr's SQL API, preventing the agent from accidentally loading thousands of records and exceeding context limits.

Claude Desktop and other MCP-aware tools can use the server to manage customers, products, invoices, orders, and contacts in a Dolibarr instance.

Consult the bundled documentation index for deep dives into configuration, API coverage, and contributor workflows.

✨ Features

• Full ERP coverage – CRUD tools for users, customers, products, invoices, orders, contacts, projects, and raw API access.

• Advanced Search – Server-side filtering for products, customers, and projects to minimize token usage and costs.

• Async/await HTTP client – Efficient Dolibarr API wrapper with structured error handling.

• Ready for MCP hosts – STDIO transport compatible with Claude Desktop out of the box.

• Shared workflow with prestashop-mcp – Identical developer ergonomics and documentation structure across both repositories.

✅ Prerequisites

• Python 3.8 or newer.

• Access to a Dolibarr installation with the REST API enabled and a personal API token.

📦 Installation

Linux / macOS

While the virtual environment is active record the Python executable path with which python. Claude Desktop must launch the MCP server using this interpreter.

Windows (PowerShell)

Run Get-Command python (or Get-Command python.exe) while the environment is activated and note the absolute path. Claude Desktop should use this interpreter inside the virtual environment, for example C:\\path\\to\\dolibarr-mcp\\venv_dolibarr\\Scripts\\python.exe.

Docker (optional)

⚙️ Configuration

Environment variables

The server reads configuration from the environment or a .env file. Both DOLIBARR_URL and DOLIBARR_SHOP_URL are accepted for the base API address.

Example .env:

Claude Desktop configuration

Add the following block to claude_desktop_config.json, replacing the paths and credentials with your own values:

Restart Claude Desktop after saving the configuration. The MCP server reads the same environment variables when launched from Linux or macOS hosts.

▶️ Usage

Start the MCP server

The server communicates over STDIO by default, so run it in the foreground from the virtual environment:

Logs are written to stderr to avoid interfering with the MCP protocol. Keep the process running while Claude Desktop is active.

HTTP streaming mode (for Open WebUI or remote MCP clients)

Enable the HTTP transport by setting MCP_TRANSPORT=http (and optionally MCP_HTTP_HOST / MCP_HTTP_PORT). This keeps the server running without STDIO and exposes the Streamable HTTP transport compatible with Open WebUI:

Then point Open WebUI’s MCP configuration at http://<host>:8080/. The MCP protocol headers (including mcp-protocol-version) are handled automatically by Open WebUI’s MCP client.

Test the Dolibarr credentials

Use the standalone connectivity check before wiring the server into an MCP host:

When the environment variables are already set, omit the overrides and run python -m dolibarr_mcp.test_connection.

🧪 Development

• Run the test-suite with pytest (see docs/development.md for coverage options and Docker helpers).

• Editable installs rely on the src/ layout and expose the dolibarr-mcp console entry point for backwards compatibility.

• The repository structure, tooling, and docs intentionally mirror prestashop-mcp to keep the companion projects aligned.

📄 License

Released under the MIT License.