How MCP Gateways Serve Context to Probabilistic Agents: Correct Answers in a Tenth of the Tool Calls

CData Labs’ benchmark argues that AI budgets should be assessed by cost per correct task rather than token consumption, incorporating both model spending and reliability. Across 810 runs involving 27 models from 12 developers completing account-health tasks spanning CRM, data warehouse, and ITSM systems, CData reported that its Connect AI Gateway achieved 83% task accuracy at $0.26 per correct task, compared with 36% accuracy and $5.67 per correct task for an open-source MCP gateway. The study attributes the difference to the AI gateway’s ability to retrieve relevant context on demand and perform cross-system joins, filtering, calculations, and ranking in a single query, whereas the pass-through gateway required models to retrieve paginated data separately and perform computation in their reasoning. CData also reported fewer tool calls, tokens, confidently incorrect read results, and unintended writes through its gateway, particularly enabling lower-cost models to complete tasks more reliably. The company provides the benchmark harness for reproduction and recommends evaluating MCP gateways based on cost per correct task, where cross-system computation occurs, and how business context is delivered.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.