root@fr:~$ connecteurs
Your agents work in your tools. Not in one more tab.
Mail, calendar, spreadsheets, ERP, CRM, document base: agents connect to what you already run, through APIs and MCP, an open standard. Nothing to migrate, nothing to replace. And every call leaves a trace in the audit log.
A connector describes the tool. The agent asks. The trace stays.
The socket is called MCP, the industry's open standard: a connector describes what a tool can do, and any governed agent can use it. One connector per tool, not one integration per tool-assistant pair.
Your tools, by family
The agent joins the tool where the work already happens.
Mail and calendar
Sorting, drafting replies, keeping the calendar: the agent works in Outlook or Gmail, and nothing leaves without your validation.
Office documents
Excel, Word, your templates: packs, letters and tables fill in your own formats, every figure linked to its source.
ERP and operations
Odoo, Sage, your business ERP: reading orders, reconciliations, preparing entries, read-only by default.
Accounting and finance
Cegid, Pennylane, Agicap, Qonto, your bank: collection, matching and reminders prepared with their supporting documents.
CRM and sales
The full client file before a meeting, follow-ups prepared with the account's history, in your CRM.
Documents and DMS
SharePoint, Drive, your document base: search that cites its sources, filing and summaries on your own documents.
A tool missing from this list is not out of reach: if it exposes an API, it is within a connector's reach, including in-house software. The workshop checks your exact configuration.
Reversibility
The connector belongs to the standard. The AI layer belongs to you.
Change a tool tomorrow: your agents learn the new connector without being rewritten. Change your AI layer: your tools' MCP connectors remain valid. Integration stops being where dependence gets built, unlike the embedded AI of US SaaS suites, billed as an add-on and tied to their platform.
Under control
Every connector is an audit surface. Not a black box.
Scope per agent
Every agent declares what it can read and where it can write. Read-only is the default; the rest is decided agent by agent.
A log line per call
Which agent, which connector, which capability, which data, what time: every tool call leaves a queryable line, exportable to your SIEM.
Validation on writes
Send, edit, publish: every outbound action waits for human validation. The agent prepares, it does not sign.
On your instance
Connectors run on your side, not through a third-party relay: your access tokens and your data pass through no intermediary platform.
Integration questions
What we get asked before plugging in.
Which tools are compatible?
Any tool that exposes an API, through an MCP connector or direct integration: mail, office documents, ERP, accounting, CRM, document base, and your in-house software. The families above cover the most common cases; your exact configuration is checked in one workshop.
Do we have to migrate our data?
No. Agents work in your tools, on your data where it lives. No migration, no replacement, no copy of your records into a third-party platform: the connector reads what its scope allows, and the trace stays with you.
What happens to the integration if we change tools?
It follows. The new tool's connector replaces the old one, your agents and their processes remain. That is the point of an open standard: integration is no longer a sunk cost at every change, nor a reason to stay.
Can an in-house application be connected?
Yes, and it is a frequent case: an MCP server is built on top of your existing API. It is bounded integration work, done with your team, and the connector then belongs to you, like the rest of the layer.
The demo plugs into one of your tools. Not into a mock-up.
In one workshop, we pick a tool, plug a connector into demo data, and you watch the request, the cited result and the log line. The rest of the estate follows the same socket.