@crbonfree/mcpBuild on your numbers

Give your AI agent secure access to your dashboard.

Connect a remote-capable agent to the hosted CrbonFree MCP server. OAuth 2.1 signs the user in, scopes every read to their organization and keeps the answer tied to the same data in the dashboard.

Remote MCP server URL

https://api.crbonfree.com/mcp
illustrative session, example numbers

10

read-only tools, none of them writes

4,500+

MCP local npm package downloads with no paid marketing

Deprecated

Crbon Labs, 2026

1

remote endpoint for every supported MCP client

1

organization scope applied to every read

What is the MCP server?

The CrbonFree remote MCP server is a hosted Model Context Protocol endpoint for browser-based and local AI agents. OAuth 2.1 authenticates the user and scopes ten read-only tools to their organizations, so an agent can discover organizations and retrieve projects, connections, receipts, statements, usage summaries, billing periods, Trust badge links, plan tier details and connection setup guidance from dashboard data.

Last checked against the product on 24 September 2026. Carbon figures follow methodology v1.2.

At a glance
Endpoint
https://api.crbonfree.com/mcp
Transport
Remote HTTP MCP
Authentication
OAuth 2.1 bearer token
Discovery
/.well-known/oauth-protected-resource/mcp
Tools
10, all reads, no writes
Organization scope
Authenticated token organization
Clients
ChatGPT, Claude website and desktop app, Codex CLI, OpenCode, Claude Code, remote-capable MCP clients

Remote-first, every tool read-only

The hosted server works with browser-based and remote-capable agents over HTTP. Its ten read-only tools discover organizations and return projects, provider connections, receipts, statements, usage summaries, billing periods, Trust badge and certificate links, plan tier details and connection setup guidance. None writes, deletes or changes billing.

  • OAuth protects every read

    Your MCP client discovers the protected resource, opens the CrbonFree sign-in and receives a bearer token for the remote MCP resource audience. The client requests and receives the scopes it needs through that OAuth flow.

  • Organization scope is automatic

    The organization in the authenticated token scopes each tool call. An agent authorized for one organization cannot retrieve data from another organization.

  • Local package remains available

    The npx package remains published for local-only clients. It is a deprecated fallback, not the recommended path, and remains available for teams that still need a local stdio server.

The same numbers as your dashboard

Every remote read is scoped to the same CrbonFree organization as the authenticated user. Your agent can discover its organizations and receive the projects, connections, receipts, statements, usage and billing periods that the dashboard can show.

See the dashboard in the sample account

What it does not do

Everything it does is a read. The remote tools discover organizations and return projects, connections, receipts, statements, usage summaries, billing periods, Trust badge links, plan tier details and connection setup guidance. None writes, deletes or changes a plan. Receipts, statements and Trust badge links are available on paid plans. The remote endpoint requires OAuth 2.1 through an MCP client, and reads are limited to the organization in the authenticated token.

How the MCP server moves your data

  1. 1

    You ask in plain language

    A question about a project, connection, receipt or statement goes to your agent as it normally would.

  2. 2

    The agent picks a tool

    It selects one of ten read-only tools for organizations, projects, connections, receipts, statements, usage summaries, billing periods, Trust badge links, plan tiers or connection setup guidance.

  3. 3

    OAuth authorizes the remote server

    The MCP client sends its OAuth bearer token to the hosted endpoint, which checks the audience and organization scope.

  4. 4

    The agent answers with the numbers

    The remote server returns organization-scoped data and the agent phrases the answer.

How to get started

In ChatGPT or Claude, no code

No code and nothing to install. Add CrbonFree to the chat app you already use, sign in once, and ask about your footprint in plain English.

ChatGPT

  1. 1In ChatGPT, open Settings and turn on Developer mode.
  2. 2Open Plugins from the sidebar, select Add, then Create MCP App.
  3. 3Name it CrbonFree, paste https://api.crbonfree.com/mcp under Connection, keep OAuth, tick the confirmation box and select Create.
  4. 4Select Continue to crbonfree, sign in with your CrbonFree email and the code we send you, then select Allow. CrbonFree is connected.

Developer mode depends on your ChatGPT plan. On Business and Enterprise, a workspace admin may need to allow it first.

Claude

  1. 1On claude.ai or the Claude desktop app, open your profile menu, then Settings, then Connectors.
  2. 2Select Add, then Add custom connector. Name it CrbonFree, paste https://api.crbonfree.com/mcp as the URL and select Continue.
  3. 3Select Connect on CrbonFree, then sign in with your CrbonFree email and the code we send you.
  4. 4Select Allow. CrbonFree is connected and ready in any new chat.

Works on every Claude plan. On Team and Enterprise, an owner adds the connector once in the organization settings.

In a coding agent

  1. 1

    Add the remote server

    Add https://api.crbonfree.com/mcp with the command for the coding agent you use.

    codex mcp add crbonfree --url https://api.crbonfree.com/mcp
    Codex CLI
    claude mcp add --transport http crbonfree https://api.crbonfree.com/mcp
    Claude Code
    opencode mcp add crbonfree --url https://api.crbonfree.com/mcp
    OpenCode
  2. 2

    Sign in

    Run the matching command to sign in to your CrbonFree account.

    codex mcp login crbonfree
    Codex CLI
    claude mcp login crbonfree
    Claude Code
    opencode mcp auth crbonfree
    OpenCode
  3. 3

    Ask a question

    Ask your agent to discover organizations, list projects, review connections, retrieve a receipt or statement, summarize usage, look up a billing period, open a Trust badge link, compare plan tiers or get connection setup guidance. The remote server calls the matching read-only tool.

Questions to ask once you connect

Copy one into ChatGPT, Claude or your coding agent. Each question is answered by a read-only tool, so the reply uses the same numbers as your dashboard.

Sustainability officer

“What was our AI carbon footprint last month, broken down by provider and model?”

get_usage_summary
Sustainability officer

“Find our latest credit retirement receipt so I can cite it in the board update.”

get_receiptsPaid plans
Sustainability officer

“Give me the link to our verified sustainability badge and certificate.”

get_trust_assetsPaid plans
Finance

“Summarize our current billing period: tokens used, emissions and status.”

get_billing_periods
Finance

“Pull our most recent footprint statement for the quarterly report.”

get_statementsPaid plans
Operations

“Which of our AI providers are connected, and does any connection need attention?”

get_connections
Operations

“List our projects and tell me which one used the most tokens.”

get_projects
Getting started

“How do I connect our AWS Bedrock account to CrbonFree?”

get_connection_setup_guide
Getting started

“Compare the CrbonFree plans. Which one includes receipts?”

list_plan_tiers

Who reaches for the MCP server

Developer in an editor

Wants to list organization projects without leaving a remote-capable coding agent.

What they get: One question runs get_projects and returns only projects in the selected organization.

Sustainability officer

Needs a carbon receipt while preparing a board update.

What they get: get_receipts returns the organization-scoped receipt data for the agent to summarize when the organization has access.

Operations lead

Needs to check provider connection status or retrieve a statement for an internal ticket.

What they get: get_connections or get_statements retrieves the matching read-only organization data.

New team member

Wants to compare plan tiers or get set up before connecting a provider.

What they get: list_plan_tiers or get_connection_setup_guide returns the plan comparison or the right setup link, without leaving the coding agent.

Which CrbonFree surface do I need?

Three of them put usage in: the dashboard through provider connections, the CLI for coding agents and the browser extension for chat sessions. The other four read the numbers back into your own code. The MCP server is highlighted.

SurfaceMeters usageReads numbers backRunsNeedsLanguage
DashboardIn the browserAn account; the sample account opens without oneNone
CLI statusOn your machineAn account; device code for headless machinesNode 20 or newer
Browser extension estimatesIn ChromeOptional, per the store listingNone
TypeScript SDKNode 18 or newerAn API keyTypeScript or JavaScript
Python SDKPython 3.8 or newerAn API keyPython
MCP serverthis pageRemote MCP endpointA CrbonFree accountAny MCP client
REST APIOver HTTPSAn API keyAny

Questions

Questions about the MCP server.

Short answers. The package page and the API reference have the long ones.

Open the dashboard
  • The remote server reads the data available to your CrbonFree organization. Usage and carbon estimates are available on the free plan, while receipts and audit packs depend on the plan.

  • Yes. The recommended endpoint is https://api.crbonfree.com/mcp. It uses remote HTTP MCP and OAuth 2.1, so remote-capable agents can connect without running a local package.

  • ChatGPT and Claude on the web or desktop connect as a custom connector with no code. Codex CLI, OpenCode and Claude Code connect with one command each. Any MCP client that supports a remote HTTP server and OAuth 2.1 can use the same endpoint.

  • No. All ten tools are reads. There are no mutations, no deletions and no billing changes, so an agent holding this server can look at your numbers but cannot alter them.

  • Your MCP client manages the OAuth token after you sign in. The remote server validates that token for its own audience and uses its organization context to scope every read.

See your own number instead of ours.

However your team already runs AI, that is how you connect. Paste a read-only provider key, drop in an SDK, add the MCP server, install the CLI or switch on the browser extension, and the first measured tokens reach your dashboard within seconds.

Create your free account

Free plan · first tokens in seconds