For maintenance teams running CARL Source

A CARL Source MCP server that answers on work orders, equipment and history under each user's own rights

A CARL Source MCP server gives Claude, ChatGPT or Copilot named tools for work orders, equipment, maintenance history and purchases. CARL names REST, OData and WebHook interfaces but publishes no reference, so every backend is confirmed against your instance. CARL ships no MCP server as of 21 September 2026.

Why CARL Source is hard for agents

Where a CARL Source connection needs engineering, not configuration

01 · Docs

No public API reference

CARL names REST, OData and WebHooks but publishes no reference. Resources and authentication have to come from CARL Berger-Levrault or from your own instance, which is the first task of the blueprint.

02 · Configuration

Every installation is parameterised

CARL Source is built for extensive parameterisation and customisation, so field names and workflows differ per customer. Tools are written against your configuration, not a generic model.

03 · Deployment

SaaS and on premises differ

Interface availability differs between SaaS and on-premises installations, and Makini reports many on-premises deployments without a usable API. Whether your installation has one decides the whole approach.

04 · Language

Documentation is French first

Documentation and much of the partner ecosystem are French first. Budget time for a French-speaking CARL administrator in the blueprint, or answers on field meaning take longer.

Checked 21 September 2026

CARL Source at a glance: interfaces, access and what exists for AI

	What we found
Interfaces	Web Services: REST API and WebHooks (REST/JSON, docs not published): CARL names configurable interfaces and web services (REST, OData, WebHook), with each application process available as a service; no public reference, SaaS or on premises · OData feed for BI (OData, docs not published): the user defines a data model in the CMMS and CARL Source generates a URL for BI tools such as Excel or Power BI; CARL names it as the way to reach SaaS data for reporting
Authentication	Not documented publicly.
Vendor MCP server	No vendor server. No MCP server from CARL Berger-Levrault found; the commercial Makini service lists a Carl Software integration with MCP.
Vendor AI assistant	CARL AI.Gen, a generative-AI maintenance helper inside CARL Source, Touch and Flash that answers everyday questions such as troubleshooting; CARL also sells CARL Predict for IoT predictive maintenance and CARL Optim for schedule and route optimisation.
Other connectors	Makini offers a commercial hosted unified API with MCP that lists Carl Software; Makini itself states that some CARL editions expose only limited or inconsistent APIs and many on-premises deployments have none.

CARL Berger-Levrault, part of the Berger-Levrault Group (FR). Maintenance organisations in industry, public transport, healthcare, property and facility management and local authorities, mainly in France and other French-speaking and European markets.

What we build

A first CARL Source tool list, written as the manifest your IT team signs

CARL Berger-Levrault, part of the Berger-Levrault Group does not publish its interface reference, so every backend below is confirmed against your installation in the blueprint before a tool is built.

# carl-source-mcp-server: tool manifest, draft for blueprint week 1
# Every tool runs as the person asking. Nothing is written without a person confirming.
tools:
  search_work_orders:      # equipment, status, date range -> work orders with priority
    backend: confirmed in blueprint
    mode: read
  get_equipment:           # equipment code -> asset record with location and status
    backend: confirmed in blueprint
    mode: read
  get_maintenance_history: # equipment, period -> past interventions and causes
    backend: OData data model, resource confirmed in blueprint
    mode: read
  list_pending_purchases:  # site -> purchases awaiting validation
    backend: confirmed in blueprint
    mode: read
  draft_work_request:      # equipment, symptom text -> draft work request for a person to submit
    backend: confirmed in blueprint
    mode: propose
audit: every call logged with user, tool, arguments and the CARL Source response

The manifest is the contract. A question that needs an operation not on it gets a new line and a review, not a wider permission.

CARL Source holds the records. An assistant without a governed way in answers from whatever someone pasted into the chat. The connector is the governed way in: named operations, the person's own permissions, every call logged.

What the connector answers, and what it will not do

Questions it answers

Which equipment is down right now, and which work orders are open on it?
What maintenance history and preventive plan exist for pump P-204?
Which purchase requests wait for validation, and which service contracts expire this quarter?

What it will not do

Create a work order by itself: the one write tool drafts a work request, and a person submits it in CARL Source.
Read more than the asking user may read in CARL Source; permissions come from CARL, not from the connector.
Replace BI reporting: long-range cost and budget analysis belongs on the OData feed and a warehouse, not in per-question calls.

When this is the wrong page

If technicians only need troubleshooting help inside CARL, CARL AI.Gen is the shorter path and CARL maintains it; if the need is reporting, the OData feed into Power BI already covers it. A connector is right when the assistant lives in Claude, ChatGPT or Copilot, or has to combine CARL with ERP or OT data.

Start with the Readiness Blueprint.

Two weeks, €4,900 fixed price. Week one agrees the tool list with your CARL Source owners and checks every backend against your release and licence. Week two tests whether a person's own CARL Source permissions reach through the connector. You keep the manifest and a costed build plan, whoever builds it. All prices are net and exclude VAT.

What we need from you: a named CARL Source administrator, a test or sandbox system we may call, and one business owner who can say which questions the assistant should answer first. The first production connector is usually a use-case pilot from €15,000; it runs in your environment, and the code is yours.

Sources: CARL Source product page · CARL Source integration · FAQ: accessing SaaS data · CARL AI solutions · CARL Source Factory · Makini CARL integration. Read 21 September 2026. Vendor documentation changes; verify against the current release.

Connectors for 100 source systems — the dated catalogue this page belongs to
Claude MCP server development — how we build an MCP server for any internal system

Questions CARL Source teams ask before the first tool call

Does CARL Source have its own MCP server?

Not as of 21 September 2026. CARL's AI is CARL AI.Gen inside the CMMS. The commercial Makini service lists a Carl Software integration with MCP, and Makini itself states that many on-premises deployments have no usable API.

Do we need extra CARL licences or access for an AI connector?

CARL publishes no API reference and no licence terms for API use. Ask CARL Berger-Levrault which web services your edition and deployment include; the answer differs between SaaS and on premises, and the blueprint starts with it.

Can the assistant see data a user is not allowed to see in CARL Source?

No. Each call runs as the asking person, so CARL's own permissions decide every answer. Two users asking the same question can get two different answers, and that is tested before go-live.

What does a CARL Source connector cost?

The two-week Readiness Blueprint is fixed at €4,900: agreed tool list, backends checked against your release, identity tested end to end, costed build plan. The production connector is usually a use-case pilot from €15,000. Prices are net, excluding VAT.

Next step

Not a sales call. An architecture call.

Thirty minutes with the engineer who would build the connector, with your CARL Source administrator in the room if you like.