Docs
Connector for AI assistants (MCP)
Ask Claude, ChatGPT, Cursor or Grok where your LLM spend went, using your own ToTheCent workspace.
Effective: 19 October 2026
What it does
The ToTheCent connector lets an AI assistant answer questions about your LLM API spend from your own workspace: how much you spent this month by provider, model and feature, how much of it ties out to the provider invoice, which calls you overpay for, and the configuration change for each fix. Every number comes from your latest ToTheCent report and keeps its label, estimated or reconciled.
The connector is read-only and returns aggregates only. It never sees prompt or completion content and it cannot change anything in your account or your configuration.
Connect
Add a custom connector (remote MCP server) in your assistant with this URL:
https://api.tothecent.ai/mcp/connector/
- Claude: Settings, Connectors, Add custom connector.
- ChatGPT: Settings, Apps and Connectors, add a custom connector (developer mode).
- Cursor: Settings, MCP, add a remote server with the URL above.
The assistant opens a ToTheCent sign-in. Sign in with your ToTheCent account and approve the assistant on the consent page. You need an account with a workspace that already has a report; if you do not have one yet, write to contact@tothecent.ai.
Tools
- Spend report (
spend_report): this month's spend by provider, model and feature, the invoice reconciliation (coverage, residual, and how far a price-list estimate would be off) and the savings available. Per-feature numbers are measured attribution, not reconciled. - Top overpaying calls (
top_overpaying_calls): the call patterns you overpay for, ranked by monthly savings, each with its fix, confidence and risk.limitis 1 to 10. Savings per row are standalone and do not add up; the total is reported separately. - Config change for a fix (
get_config_diff): the before-to-after configuration change for one fix, for your stack if you name it (for examplelitellm). You apply it; ToTheCent never does.
If your account belongs to several workspaces, the assistant passesworkspace (its slug) to choose one. Each answer states which workspace and which report snapshot it comes from.
Limits and privacy
The free plan includes 300 connector calls per workspace per calendar month (UTC). Above that, a tool returns a message with a link to upgrade instead of data. The connector's data handling is described in section 7 of the Privacy Policy. To disconnect, remove the connector in your assistant's settings.