Announcing the Muir MCP Connection: Product Intelligence Inside Your Enterprise AI

Today we are announcing the Muir MCP connection: a native integration that puts Muir's product intelligence, from teardown-grade cost models to multi-tier supply chain maps and ISO-grade carbon data, inside the AI assistants your team already uses. Connect Muir to your enterprise LLM and build products natively in chat, explore the full granularity of every model, and put agentic capabilities to work building new product strategies.

Scaling the impact of Muir

The assumptions and modeling behind a single product model mean thousands of outputs. These are broad data points spanning material selection, mass, manufacturing process, country of origin, labor rates, and electricity consumption, to name a few. This meant that even with scaled product modeling, users were not always able to scale their decision making. So much goes into the making of one product — its cost, supply chain, and emissions — that no one individual can easily conceptualize and act on that information.

The Model Context Protocol (MCP) changes this. MCP lets an AI assistant call Muir's platform for you: not a summary of your data, not a copy of it, but the live model, with the assistant translating between plain language and the analytical machinery underneath. Muir's MCP connection exposes the full workflow, building models, interrogating them, editing them, and running scenarios, to any assistant your enterprise has approved.

The result is that the full power of Muir is no longer limited by clicks, or by any one person's ability to review thousands of data points.

What the video shows, beat by beat

A product model from a single image. The workflow opens with nothing but a photo. Our user asks the assistant to create a new product model from it, and Muir's synthetic BoM engine decomposes the product into its component hierarchy: parts, subparts, materials, masses, manufacturing processes, and prices, each line carrying its derivation. What used to require forty hours of data cleanup can now be done from a picture.

Interrogating the model at full depth. Our user then asks a question that should be easy but isn't across large BoMs: how much steel is used in this product? Behind the conversation, the assistant traverses the full component tree, aggregates mass by material, and returns the answer with the line items behind it. The model always contained this answer. What changed is the ease of reaching it.

Expert edits, not just questions. The connection is not a read-only reporting layer. Midway through, our user sees an incorrect assumption and overrides the inferred material with actuals. Domain expertise stays in the loop: the assistant handles the mechanics of the change, and the human supplies the judgment about what should change.

Impact before commitment. Nothing is final without approval. Before the edit lands, the assistant surfaces its full impact, what moves in cost, mass, and emissions, and our user applies it explicitly. Every change is proposed, validated against the model, previewed, and only then committed, with attribution. This is the property that makes the output usable in a negotiation rather than a black box nobody trusts.

From interrogation to strategy. With the model current, our user hands the agent a strategy problem: where are the cost reductions? The assistant reviews the full output, every material, process, and sourcing decision in the build, and returns ranked levers drawn from the product's actual value chain. Our user goes deeper on the one that matters most: switching suppliers on a key component, with the cost and sourcing consequences quantified in the same conversation. This is the second thing the connection adds. Not just answers about the model, but an agent that can review the whole of it and recommend what to do.

Institutional memory at the moment of decision. Before the decision gets made, the assistant surfaces something else: a teammate ran a similar exploration days earlier. Because the connection operates on the shared platform rather than a private chat session, prior teamwork appears when it can still shape the decision instead of six months later in a postmortem. Our user picks up the team's analysis, applies the team's decision, and creates the scenario.

The conversation ends in an artifact. The output is not a chat transcript. It is a saved scenario in Muir, versioned, attributed, and visible to the rest of the team, ready to carry into a design review or a supplier negotiation.

Scaling users to meet their product portfolio

Every Muir model has always held more than any person reads. A full teardown carries the material, mass, process, origin, price, and emissions basis for every node in the hierarchy, thousands of data points on a single product, multiplied across a portfolio. The questions that cut across that structure, how much of one material sits in the whole build, which lines share an upstream supplier, were answerable in principle and rarely asked in practice, because each one cost hours of manual traversal.

With our MCP connection, any question can now be asked. A design engineer can price a material or geometry change at the moment of selection, while it can still change the outcome. A buyer can pull a negotiation-grade should-cost position on any line in the BoM, not just the lines someone had time to study. A sustainability lead can query the carbon basis of the same model, in the same conversation, as the cost questions. None of these are new capabilities of the model. They are questions the model could always answer, finally easy enough to ask.

An agent on top of the output

The deeper change is what happens after a question is answered. The assistant is given the same ability to leverage Muir's models as a user. Agents can review any model, test hypotheses, rank the cost-reduction levers by impact, quantify the supplier switch, check whether the team has already explored the same direction, and assemble the result into a scenario the organization can act on.

The user has final say, choosing which lever to pull and supplying the expert edits the model cannot infer, while the agent handles the breadth no person can. That accelerates a shift we have written about before: from producing the number to deciding what the company does with it. The difference now is that the number gets produced across the entire model at once, with the reasoning attached.

Built for enterprise governance from the start

Your product data contains critical intellectual property, making security essential. Importantly, your product data stays with Muir; the MCP connection is a permissioned interface to the platform, not an export of it, and it inherits the access controls your workspace already enforces. Every value in a Muir model carries provenance, so the assistant's answers trace to their basis: what the source declared, what the system generated, what a human edited. Every write follows the propose-validate-preview-apply path shown in the video, so no model changes without an explicit, attributed decision. And the platform behind the connection is SOC 2 Type II certified.

The assistant is a new interface. The rigor underneath it is unchanged.

Beta available now

The Muir MCP connection is available to select Muir customers today. Setup takes minutes: connect the Muir MCP server in your assistant's integrations, authenticate against your Muir workspace, and every model your team has built is in the conversation.

If you want to see it live on one of your own products, from a photo or a raw BoM to a sourcing scenario in a single conversation, book a demo.