Introducing agent-native PCF: calculate product carbon footprints through our new MCP

Key Takeaways

‍‍

Our PCF platform is now an MCP server: your AI agent can calculate a real footprint, not just talk about.
‍

An auditable MCP: every result passes a safety check, with source and confidence on every value, so you get a defensible number, not a guess.
‍

It plugs into the agents and systems you already run, with no new tool to learn.

‍

‍

We are reaching a huge milestone in embedding our carbon engine into AI and agentic workflows: Our PCF platform is available as a Model Context Protocol (MCP) server, so the agent you already work with (Claude Cowork, a custom internal agent, or any MCP-compatible tool) can upload a bill of materials, run the calculation, interpret the result, model decarbonisation scenarios, and draft an ISO 14067 or PACT Pathfinder report. The footprint comes to you, inside the conversation you are already having, instead of sending you to another tab. It is available now: install our PCF plugin in Cowork or connect via MCP. And it is auditable: every value carries its source and confidence.

‍

PCF has always carried a high barrier. The methodology choices are technical, bills of materials sprawl across hundreds of line items, emission-factor databases are opaque, and compliance requirements sit on top of all of it. So the work usually means a specialist, several sessions, and a lot of clicking through a web app. Agent-native PCF removes that friction by bringing the calculation to where the work already happens. A multi-tab, multi-session process becomes a conversation.
‍

What the MCP gives you

The PCF MCP connects an AI agent directly to our platform through a structured set of tools. The agent can create and manage products, upload bills of materials, configure methodology settings, trigger calculations, perform product portfolio analytics, retrieve results, search the emission-factor database, and assemble reports, all without you leaving your workflow. You describe what you need in plain language, and the agent requests only the specific data the calculation still needs, rather than handing you a long upfront form.

‍

Three examples show the range:

‍

• A sustainability manager describes a product in plain language, points the agent at the bill of materials, and has a footprint running in minutes, then asks where the emissions concentrate.
  ‍
• A procurement analyst asks "what happens to our footprint if we switch to recycled aluminium?" and gets a modelled before-and-after against a verified baseline, sized before the sourcing conversation starts.
  ‍
• A consultant generates a formatted ISO 14067 report draft directly from a calculated assessment, with data quality and boundary exclusions surfaced up front, ready for client review.

‍

In each case the expertise travels with the request, so the result is something you can stand behind rather than a number you have to second-guess.
‍

The skills layer, and why it matters

Raw tool access is only half the picture. The value sits in the skills layered on top of the MCP, which encode LCA expertise into each step. Each skill knows what to ask, what tends to go wrong, and what to check before moving on. Seven skills cover the workflow:

‍

• BoM Upload and Validation catches structural issues, unit mismatches, and missing material descriptions before they cause a silent import failure.
  ‍
• Methodology Setup walks through the choices that shape the result (ISO 14067 vs PACT, allocation method, system boundary, cut-off rules) and writes them to the assessment, so they travel with the result for audit purposes.
  ‍
• Pre-Calc Gate intercepts the calculation before it runs and surfaces every input the platform would otherwise default (energy assumptions, output unit, emission-factor mapping on hotspots), so you confirm or override each one deliberately.
  ‍
• Result Safety Check independently validates a calculated footprint before anyone sees it, checking mass coverage, order-of-magnitude plausibility, emission-factor provenance, and known platform edge cases.
  ‍
• Result Interpretation explains the number in plain language, including where emissions come from, which materials and suppliers dominate, and the highest-leverage reduction opportunities.
  ‍
• Scenario Modelling sizes decarbonisation levers, such as switching to green steel, against a verified baseline, without committing fabricated data to a saved assessment.
  ‍
• Report Draft assembles a formatted, disclosure-first report as PDF and DOCX, built around the platform's ISO 14067 export, with data quality, primary-data share, and boundary exclusions surfaced first.

‍

The safety check is the difference that matters. Every result passes an independent validation before it is shared, so what you present is a defensible footprint, not a confident-looking guess. That validation step is exactly what low-cost AI estimators skip, which is why their output reads convincingly while quietly being wrong. With the safety check in place, you can take the number into a customer meeting or an audit and explain how it was reached. That is what makes this an auditable MCP rather than a black box: every value carries its source and confidence, so the footprint holds up when someone checks the working.
‍

What still needs your review

Agent-native does not mean unattended. The Pre-Calc Gate exists so you confirm or knowingly skip each defaulted input, and the Result Interpretation step is there for you to read, not to rubber-stamp. Estimated values are labelled as estimates and can be improved over time as primary data arrives. The right workflow depends on your product data, BoM structure, and customer requirements, so treat the agent as a fast, expert colleague whose work you still sign off on. That review is what keeps the footprint yours and defensible.
‍

Integrating into your own systems

For partners and developers the priority is different: adding carbon management to what you already offer, opening new revenue, and getting there without a long build. The MCP is designed for that. Here is how it composes:

‍

• Internal agents can call the PCF MCP as a tool. A procurement agent can trigger a PCF comparison when it evaluates a material switch, and a reporting agent can pull verified footprints straight into quarterly ESG output without a manual export step.
  ‍
• Claude Cowork (or a similar AI workspace) needs no integration work. Install the plugin and the skills are available in existing sessions immediately. Ask a question, get a footprint.
  ‍
• Developers and custom workflows get a structured tool interface, so PCF capabilities compose with the systems you already run. Emission-factor search, BoM upload, and the calculation tools are callable individually, which means integration can be as narrow or as broad as the workflow demands, alongside inventory systems, ERP data, and supplier portals.
  ‍
• The skills layer is portable. Because the skills encode workflow logic and quality-assurance checks rather than raw computation, you can adapt them for specific industries, products, or reporting requirements without rebuilding the platform connection.

‍

This answers a clear demand signal. Enterprise carbon and procurement platforms are already asking for a way to give their own customers real supplier PCFs without sending those suppliers off to a separate tool. The MCP lets you offer a real, auditable footprint inside the product your customers already use, which is a faster route to market than building a calculation engine from scratch.
‍

From friction to fluency

The point of agent-native PCF is to put expertise where teams already work. A sustainability manager should not have to become an LCA specialist to produce a defensible number, and a procurement team should not have to leave its tools to ask a carbon question. By delivering the engine and its built-in checks through the agent people already use, we turn carbon footprinting from a specialist project into a conversation that ends in a number you can stand behind. One connection covers every step from BoM to boardroom-ready report, checked before it reaches you. That is the shift from friction to fluency, and it is available to use today.
‍

FAQ

Can an AI agent really calculate a product carbon footprint?
‍

Yes. Through our MCP, the agent uploads the bill of materials, runs the calculation on our PCF engine, and returns a result with source and confidence visible on every value. It does the work, rather than answering questions about how it would be done.
‍

Is the output compliant with standards?
‍

The PCF is compliant with ISO 14067:2018 and PACT v3.0, and aligned with the GHG Protocol Product Standard and the Catena-X PCF Rulebook v4.0. It is compliant, not yet third-party certified.

‍

How is this different from other AI carbon estimators?

‍

Every result passes an independent safety check before it is shared, covering mass coverage, order-of-magnitude plausibility, and emission-factor provenance. That validation is what makes the footprint defensible rather than a guess that happens to look confident. Hence you can ensure full auditability of the result according to expert requirements.

‍

How do I get a PDF or DOCX report?
‍

The Report Draft skill assembles the formatted PDF and DOCX, built around the platform's ISO 14067 export. The platform's own native exports are the ISO 14067 Excel report and the PACT-compliant CSV.

‍

What do I need to start?
‍

A product and its bill of materials, plus the agent you already work  with. Install our PCF plugin in Cowork or connect via MCP, and the agent requests the rest as the calculation needs it.‍
‍

Get the job done, with your AI agent

Agent-native PCF takes the footprint to where you already work, runs it through expert checks, and gives you a result you can defend. The PCF MCP and plugin are available now. Install the forward earth PCF plugin in Cowork or connect via MCP, and run your first footprint from BoM to report inside one conversation. To see it on your own product data, book a PCF demo.

‍