Ask a fleet manager where their data lives and the answer is usually a dashboard. Vehicle locations, driver hours, safety events, fuel levels and inspection reports sit behind menus, filters and reports that someone has to open, configure and read.
Samsara, one of the largest providers of connected operations software for fleets and physical operations, is testing a different route. Through a service built on the Model Context Protocol, known as MCP, customers can connect AI assistants directly to their Samsara data and ask questions in plain language. Which trucks are below a quarter tank? Which drivers still have four hours of drive time left? Which trailers have been sitting at a customer’s yard for more than two days?
On its own, that sounds like a convenience feature. Look more closely and it signals a broader shift in how IoT data may be reached and used, and it raises questions that every platform connecting physical operations to AI will have to answer.
What MCP Is
To understand what Samsara has built, it helps to understand the protocol underneath it.
A Common Plug For AI Tools
The Model Context Protocol is an open standard for connecting AI assistants to outside systems, introduced by Anthropic in late 2024. Samsara’s own documentation describes it as “the open standard that tools like Claude use to talk to external systems.”
Before standards like MCP, connecting an AI assistant to a business system typically meant building a custom integration for each combination of assistant and data source. MCP defines a common way for a system to describe the actions and queries it offers, and for an AI client to discover and use them. A company that builds one MCP server can, in principle, make its data available to any AI tool that supports the protocol.
Why It Spread
The appeal is practical. Businesses are adopting several AI tools at once, and software vendors do not want to build and maintain separate connectors for each. A shared protocol reduces that work on both sides. Over the past two years, MCP support has appeared across a wide range of AI assistants, developer tools and business software.
For IoT platforms, which hold large volumes of operational data behind their own interfaces, that creates an obvious opportunity: let customers reach that data from whichever AI assistant they already use.
What Samsara Has Built
Samsara’s implementation is documented on its developer portal’s MCP page. It describes the service as a way to “connect AI assistants and agents directly to your Samsara data.”
One Server, Three Regions
The service runs as a remote MCP server, with separate endpoints for the United States, Europe and the UK, and Canada. Customers must use the endpoint that matches the region of their Samsara account. The documentation lists Claude Desktop and the Cursor coding tool as supported clients, along with any client that supports remote MCP servers with OAuth sign-in.
Setup is designed to be light. In Claude Desktop, an organisation owner adds Samsara as a custom connector, and users then enable it in their own settings. The first time a client connects, it opens a browser window where the user signs in with their Samsara credentials.
Read-Only By Design
The most important design decision is a limit. According to Samsara, the MCP service “can answer questions about your data but cannot change anything in your Samsara account.” Its tools reference is equally direct: “No tool in this catalog can modify your Samsara data.”
That means an AI assistant connected through MCP can look up vehicles, drivers, safety events or inspection reports, but cannot dispatch a truck, change a route, edit a driver’s record or close a maintenance ticket. Where a workflow requires a change, the documentation describes handing off to other systems, such as maintenance or service management tools.
Permissions That Follow The User
The second key decision concerns access. Samsara says the MCP server “enforces the same identity, role, and tag scoping as the Samsara dashboard.” In practice, an administrator sees administrator-level data, while a user limited to certain tags, such as a region or crew, sees only those vehicles and drivers through the AI assistant as well.
This is important because AI assistants make data easy to ask for. If permissions were looser in the AI channel than in the dashboard, a single question could expose information a user is not meant to see.
Status
Samsara’s main MCP page, updated on 16 September 2026, describes the service as “generally available for supported Samsara organizations.” Its tools reference page, however, still describes the service as a “closed, read-only beta” and notes that tool names and capabilities may change. Customers interested in using it should confirm availability for their own account with Samsara.
What You Can Ask It
The tools reference lists the queries the MCP service supports, grouped by the kind of work they cover. Samsara’s documentation index refers to a catalogue of 44 read-only queries.
Fleet And Assets
The fleet tools find vehicles, drivers, assets and trailers by name, list current locations and status, and read engine gauges and fuel levels. Example prompts include “Find the vehicle called Unit 118” and “Which trucks on the Red Crew tag are below a quarter tank?”
Other tools cover how long trailers and assets have been sitting at yards or customer sites, and which trailers are being held past an agreed grace period, known as detention. One example: “Which trailers are currently in detention, and at which customers?”
Safety And Compliance
Safety tools return events such as harsh braking, filtered by type, severity and time, and analyse speeding by vehicle. Compliance tools report hours-of-service status, including how much driving and duty time each driver has left. An example prompt: “Which drivers have at least 4 hours of drive time remaining right now?”
Maintenance And Inspections
The maintenance tools list driver vehicle inspection reports, known as DVIRs, show individual inspection details and list defects with their resolution status. Samsara gives the example “List all unresolved defects from this week’s DVIRs.”
Locations And Tags
Two further tools give the assistant context. One returns the organisation’s saved places and geofences, such as yards, terminals and customer sites. The other returns its tag structure, the way a company groups vehicles, drivers and assets into crews, regions or business units.
These tools matter because most real questions are framed in a company’s own terms. A dispatcher is more likely to ask about “the Blue Crew” or “the northern depot” than about vehicle identifiers. By reading the organisation’s own tags and places, the assistant can translate everyday language into the right filters, and return answers that match how the business is actually organised.
The documentation also notes a current gap: safety events cannot yet be filtered by location.
Where It Fits In Samsara’s AI Plans
MCP is one part of a broader push by Samsara into AI features.
In June 2026, the company announced Agent Studio, a no-code environment for building AI agents that monitor operations, follow company policies and take defined actions within set guardrails. It launched with more than 15 templates and was offered in early access for safety and maintenance workflows. Samsara says its network collects 25 trillion data points a year.
In its September product updates, Samsara added its AI assistant to the driver app, giving drivers answers about remaining drive time, hours-of-service violations, safety scores and route details, as well as how-to guidance. It also highlighted technician performance reporting through natural-language questions and new maintenance API endpoints for syncing with ERP systems.
The pieces fit together. Samsara’s own assistant and Agent Studio work inside its platform. MCP opens the same data to AI tools outside it. The REST API remains the route for heavier integration: Samsara’s documentation is explicit that “the REST API remains the right tool for building integrations and moving data at volume.”
Why It Matters For IoT
Samsara is not the only IoT platform exploring AI access, but its approach illustrates several trends likely to spread.
Data Access Without Dashboards
For years, the main interface to IoT data has been the dashboard. Dashboards are powerful but rigid: someone has to decide in advance what to show, and users have to know where to look. Natural-language access through an AI assistant changes that. A dispatcher can ask a specific question when it comes up, in words rather than filters.
That matters most for occasional or cross-cutting questions that no dashboard was built to answer, such as combining fuel levels, driver hours and tomorrow’s schedule in a single request.
A Pattern Other Platforms Will Follow
Connected operations data is spread across many platforms: telematics, building management, industrial monitoring, asset tracking. If each exposes a standard MCP server, a single AI assistant could, in principle, answer questions that span several of them. That is a significant change from today, where each platform keeps its data behind its own interface.
We traced how connectivity and data reshaped vehicles in how IoT rewired the car. AI access through standard protocols is a next layer on top of that data.
The Read-Only Line
Samsara’s decision to keep MCP read-only is worth noting. Letting an AI assistant read data carries risks, but letting it change operational systems carries far greater ones. A mistaken query produces a wrong answer. A mistaken action could reroute a truck or clear a safety defect.
For now, Samsara keeps actions inside its own platform, where Agent Studio operates with defined triggers and guardrails, and keeps the open MCP channel limited to questions. Whether and how that line moves will be one of the more important decisions for platforms of this kind.
What To Watch Out For
The convenience of asking questions in plain language brings responsibilities for both vendors and customers.
The first is accuracy. AI assistants can misread a question, choose the wrong tool or summarise results incorrectly. For operational decisions, users should check important answers against the source data, at least until they understand how reliable the assistant is for their questions.
The second is where data goes. When an AI assistant retrieves Samsara data, that data passes to the AI tool the user is running. Organisations should understand the data handling and retention terms of the AI services they connect, not only those of Samsara.
The third is governance. Deciding who in an organisation can connect AI clients, and which clients are approved, becomes part of managing access to operational data.
Questions For Fleet Operators
For companies running Samsara or similar platforms, a few questions can guide whether and how to use AI access.
Which questions do your teams ask repeatedly that dashboards answer poorly? Those are the best candidates for natural-language access.
Which AI tools are approved in your organisation, and do their data terms fit your policies for operational and driver data?
Are role and tag permissions in your platform set up correctly? With AI access, any existing gaps in permissions become easier to expose.
How will you check that answers are correct before acting on them, particularly for safety and compliance?
A Question-Answering Layer For Physical Operations
Samsara’s MCP service is, for now, a modest step: a read-only way for AI assistants to answer questions about vehicles, drivers, safety and maintenance. Its significance lies less in what it does today than in the pattern it represents.
IoT platforms have spent a decade collecting data about physical operations and presenting it through their own screens. Standard protocols like MCP make that data reachable from wherever people already work, in the words they already use. How far that goes, and when read-only access gives way to carefully controlled actions, will shape how connected operations are run in the years ahead.
