TokenOracle
Wire hosted MCP tools into Claude or Cursor so agents estimate token cost, compare models, and route calls against a budget before you burn credits.
Overview
TokenOracle is a MCP server for the Operate phase that estimates LLM token costs, compares models, and supports budget-aware routing for agent workloads.
What is this MCP server?
- Hosted streamable-http MCP at mcp.guffeyholdings.com/TokenOracle (v1.0.1) plus npm stdio package token-oracle-mcp
- LLM cost estimation and side-by-side model comparison from catalog metadata
- Budget-aware routing guidance so agents pick cheaper or fit-for-task models
- Required X-API-Key / TOKEN_ORACLE_API_KEY with optional TOKEN_ORACLE_SUBJECT for per-user attribution
- Optional TOKEN_ORACLE_BASE_URL override for non-default hosted endpoints
- Server version 1.0.1
- npm package identifier token-oracle-mcp with stdio transport
- Default hosted endpoint https://mcp.guffeyholdings.com/TokenOracle
What problem does it solve?
You cannot see or cap model spend inside the agent loop, so multi-step coding sessions rack up surprise bills across providers.
Who is it for?
Indie builders running frequent agent sessions who want hosted MCP cost tools without building their own pricing spreadsheet.
Skip if: Teams that only use a single fixed model with flat pricing and never compare alternatives or track per-feature spend.
What do I get? / Deliverables
After you register TokenOracle, your assistant can quote costs and compare models before routing, so routing choices align with your budget.
- Registered TokenOracle MCP connection in your agent config
- Cost estimates and model comparisons usable inside tool calls
- Optional per-subject attribution via X-Token-Oracle-Subject or TOKEN_ORACLE_SUBJECT
Recommended MCP Servers
Journey fit
Recurring LLM spend and routing decisions are production operations—solo builders need cost control after they ship agent features, not only during initial integration. Infra is the canonical shelf because TokenOracle sits on the hosted MCP endpoint and environment variables that govern how workloads hit models in production.
How it compares
Hosted MCP cost oracle for agents, not a generic LLM proxy or local tokenizer CLI.
Common Questions / FAQ
Who is TokenOracle for?
Solo builders and small teams using Claude Code, Cursor, or Codex who need MCP-native cost estimates and model comparisons while shipping agent features.
When should I use TokenOracle?
Use it before heavy agent runs, when choosing between models for a feature, or while operating production agents where monthly LLM spend must stay predictable.
How do I add TokenOracle to my agent?
Add the streamable-http remote https://mcp.guffeyholdings.com/TokenOracle with header X-API-Key, or install npm token-oracle-mcp and set TOKEN_ORACLE_API_KEY (optional TOKEN_ORACLE_BASE_URL and TOKEN_ORACLE_SUBJECT).