Task management
Microsoft To Do, Notion, Trello, Asana
AI creates, moves, completes your tasks from a conversation. Ideal for synthesizing a meeting into a TODO list, or letting AI prioritize your week.
MCP (Model Context Protocol) is the open standard that lets Claude, ChatGPT, Cursor and other LLMs act on your systems — not just chat. We build the custom connector you need: secure, production-ready.
Définition. The Model Context Protocol (MCP) is an open standard published by Anthropic in November 2024. It enables an AI assistant (Claude, ChatGPT, Cursor, Continue…) to access tools, data and actions through an intermediary server you control. An MCP exposes three primitives: tools (executable actions), resources (readable data) and prompts (reusable templates).
An MCP turns your CRM, ERP, Drive or in-house API into tools the AI can read and manipulate in natural language — with fine-grained control over what it can do.
No lock-in. Compatible with all modern AI clients (Claude, ChatGPT, Cursor…).
Your data stays with you. No vendor dependency, on-premise deployment possible.
You decide what AI can read, write or trigger. Built-in audit logs.
OAuth, monitoring, versioning, error handling. Not a POC: a service that scales.
Your team uses Claude (or another AI assistant), and you want it to act on your internal tools.
Your customers increasingly use conversational AI. Exposing your SaaS via MCP means a new interface and a new distribution channel.
Note de positionnement. MCP isn't just a technical integration: it's a product decision. We help you arbitrate what to expose, how, and at what granularity — so AI is truly useful without becoming a risk.
We qualify before quoting. We scope before coding. We deliver before billing the run.
15 minutes online to identify the tools to connect, target AI use cases and technical constraints.
Analysis of your API or target system, selection of tools and resources, choice of authentication, identification of safeguards.
Fixed scope, listed deliverables, delivery schedule and support/run conditions.
Iterative development with weekly demos, real testing with your team, usage and technical documentation, deployment.
MCP server hosting and monitoring, security and protocol compatibility updates, new tools, support and 30-day warranty included.
Fixed price for simple MCPs, quote for complex cases. Prices shown excl. VAT and incl. VAT (French VAT 20 %) — EU VAT number: FR47519015713.
For SMBs and mid-market companies who want to connect their internal tools to AI.
For SaaS publishers who want to expose their product via MCP.
Coûts non inclus. hosting (if self-managed by the client), AI service subscriptions, possible third-party fees (auth gateways, external services). Estimates communicated in the detailed quote.
Microsoft To Do, Notion, Trello, Asana
AI creates, moves, completes your tasks from a conversation. Ideal for synthesizing a meeting into a TODO list, or letting AI prioritize your week.
HubSpot, Pipedrive, Salesforce
Read opportunities, update records, generate follow-ups. AI becomes a sales assistant that acts, not just suggests.
Abby, Sellsy, Pennylane, Stripe
Create quotes and invoices from a discussion, track payments, automatically generate customer reminders.
Drive, SharePoint, Notion, Confluence
Search, read, create and update documents. AI accesses your knowledge base without copy or external indexing.
Custom connectors
For your proprietary systems. Natural-language queries, report generation, smart alerts.
Multi-tool orchestration
Signed quote → invoice → Slack notification → Notion task. AI orchestrates multiple tools in one natural workflow.
To demonstrate the method, I open-sourced a complete MCP connector for Microsoft To Do — available on GitHub. OAuth authentication, full CRUD on tasks and lists, ready to plug into Claude Desktop or claude.ai.
This project illustrates exactly what's delivered in an MCP Standard tier: clean code, install docs, OAuth safeguards. Your scope depends on your needs — we discuss it during the free scoping.
Deployed and operational, production-ready.
Git repo, README, architecture doc, diagrams.
How to use the MCP from Claude, concrete examples.
Permissions, secrets, audit logs, safeguards.
MCP (Model Context Protocol) is an open standard created by Anthropic in November 2024. It lets an AI assistant (Claude, ChatGPT, Cursor…) access external tools, data and actions — without hardcoding integration on the LLM side. Concretely, an MCP server exposes tools (actions), resources (reads) and prompts (templates) that AI can use on demand, under your control.
Yes — that's actually one of MCP's main benefits. The server can be hosted by you (on-premise or private cloud), it controls what AI can read and write, and exchanges can be logged. No copying, no third-party indexing. Security scope is part of the initial scoping.
An MCP Starter: 1 to 2 weeks. An MCP Standard: 2 to 4 weeks. A complex custom one: quoted on a case-by-case basis. In all cases, weekly demos and iterative delivery — you see the result being built.
All those that implement the MCP protocol: Claude (Desktop, Web, API), ChatGPT (via Apps), Cursor, Continue, and the list is growing rapidly. That's the advantage of the open standard: an MCP built today will work tomorrow with new clients that adopt it.
We adapt. SOAP, GraphQL, direct database, files to parse, scraping… that's precisely what distinguishes a custom MCP from a Starter. The initial scoping (free) is used to assess feasibility and quote the effort.
Yes. The source code is yours, and either you maintain it yourself, or you subscribe to the Run & maintenance offer that includes security updates, protocol compatibility, and adding new tools as needed.
Custom software development can fall under the French CII (Crédit d'Impôt Innovation) or certain regional digital transformation grants. I'm not a tax expert, but I gladly direct you to the right interlocutors.
The preliminary questionnaire takes 15 minutes and commits to nothing. Next, a free 1h technical scoping to qualify your need.