Skip to main content
Glama
Sellsy MCP Community

Sellsy MCP Community

by bouchecousue

Sellsy MCP Community

Sellsy MCP Community — Connect your CRM to AI

A self-hosted Model Context Protocol server for Sellsy API v2. Connect an MCP client to your CRM using each person's own Sellsy identity.

Community beta. Maintained by BoucheCousue; not an official Sellsy product and not affiliated with or endorsed by Sellsy. Présentation en français.

What it does

  • 42 tools for companies, contacts, items, estimates, invoices, opportunities, subscriptions, related documents, comments and files.

  • Native search, bounded pagination and explicit incomplete-result reporting.

  • Sales history with document-line evidence and optional purchase-cost previews.

  • Optional estimate and opportunity writes through separate prepare / apply tools: per-user manifests, expiry, concurrency checks and readback verification.

  • OAuth with PKCE, separate MCP and Sellsy tokens, encrypted upstream credentials, rotating MCP refresh tokens and local administration commands.

  • Shared read throttling after rate limits; business writes are not automatically retried.

Writes are disabled by default. This release is intended for one organization per deployment, with multiple individual staff accounts. It is not an audited public multi-organization hosting service.

Related MCP server: pipedrive-mcp

Quick start

Requirements: Python 3.11+ and uv. Create your own Sellsy Private OAuth application with authorization-code flow, PKCE, refresh tokens, the scopes listed in .env.example and your exact callback URL. Follow the Sellsy setup walkthrough or the guide pas à pas en français for the developer portal, scope selection and first connection.

The server uses the official MCP Python SDK 2.3+ (MCPServer). See the SDK v2 migration notes when upgrading an existing instance.

git clone https://github.com/bouchecousue/sellsy-mcp.git
cd sellsy-mcp
cp .env.example .env
chmod 600 .env
# Edit .env: credentials, HTTPS public origin, callback allowlist.
uv sync --frozen --no-dev
uv run --frozen --no-dev sellsy-mcp

Use a reverse proxy or tunnel for TLS. Forward the whole origin, including OAuth and discovery routes, to 127.0.0.1:8766. The callback is:

https://YOUR-MCP-HOST/oauth/callback

The MCP endpoint is https://YOUR-MCP-HOST/mcp. Anonymous calls return HTTP 401 with OAuth discovery metadata. /health/ready checks the service and storage; use the authenticated sellsy_get_capabilities tool to check actual Sellsy access.

For containers, configure .env first, then run:

docker compose up --build -d
docker compose exec sellsy-mcp sellsy-mcp-admin grants

The Compose example publishes loopback only. See deployment for Docker, Linux/systemd, persistence, backups and reverse-proxy requirements.

Connecting an MCP client

Use a client supporting remote Streamable HTTP and OAuth discovery/dynamic client registration. Add your HTTPS /mcp endpoint, sign in to Sellsy, then inspect sellsy_get_capabilities. Tool descriptions and most operational messages are in French.

The original integration has been exercised with ChatGPT. This community beta's tests simulate OAuth and business HTTP; they do not certify every client or every Sellsy plan. See configuration and OAuth for callback allowlists and reconnecting after scope changes.

Enabling writes

  1. Set SELLSY_MCP_ENABLE_WRITES=true and restart your instance.

  2. Reconnect the MCP client and approve both sellsy.write and the relevant Sellsy scopes (estimates.write / opportunities.write).

  3. Request prepare_*, review its diff, explicitly approve it, then request the corresponding apply_* with approved=true.

  4. Inspect the persisted result and readback. Never replay an uncertain write.

The calling application must collect the user's approval. An approved boolean is an API contract, not an independent human-approval interface or sandbox.

Documentation

License and attribution

MIT. Copyright 2026 BoucheCousue and contributors. This repository starts with a clean source snapshot; it contains no production state or private deployment history. Dependencies retain their own licenses; see third-party notices.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    C
    maintenance
    Enables AI agents to read and write CRM data—companies, contacts, opportunities, tasks, and interactions—through MCP, with OAuth and provenance tracking for every value.
    MIT
  • A
    license
    Not graded
    quality
    B
    maintenance
    Enables reading and managing Pipedrive CRM records—deals, people, organizations, activities, and notes—over MCP, including searching, writing, and moving deals through pipelines.
    Apache 2.0
  • A
    license
    Not graded
    quality
    C
    maintenance
    Enables MCP-compatible AI clients to search and view CRM customers, deals, contracts, tasks, calendars, and outbound jobs, and to propose opt-in write actions that require human approval in the CRM.
    Apache 2.0