Reasoning Step Word Count Analyzer

Reasoning Step Word Count Analyzer MCP Connector for Claude

A+

Analyzes ReAct traces to measure reasoning depth and identify zero-shot behavior.

3 tools Official Updated Oct 1, 2026 Official Vinkius Partner

This MCP server provides tools to evaluate the quality of the ReAct (Reasoning + Acting) framework. It calculates the verbosity ratio between the 'Thought' and 'Action' phases of an LLM trace. By using analyze_trace_verbosity, you can detect zero-shot behavior where the thought block is too brief. You can also use aggregate_trace_metrics to summarize reasoning patterns across entire traces or identify_high_verbosity_steps to find moments of deep cognitive effort.

reactreasoningverbosityllm-metricstrace-analysis

3 tools expose this connector's capabilities to your AI agent.

aggregate_trace_metrics

analyze_trace_verbosity

identify_high_verbosity_steps

See how to talk to your AI agent using Reasoning Step Word Count Analyzer.

Analyze this ReAct step: Thought: 'I need to find the weather.' Action: 'get_weather(city="London")'

{"thoughtWordCount": 6, "actionWordCount": 3, "verbosityRatio": 2.0, "isZeroShot": true}

Summarize these steps: [{'thoughtWordCount': 20, 'actionWordCount': 5, 'verbosityRatio': 4.0, 'isZeroShot': false}]

{"totalSteps": 1, "averageVerbosityRatio": 4.0, "zeroShotCount": 0, "totalThoughtWords": 20, "totalActionWords": 5}

Find steps with a verbosity ratio higher than 2.0 in this list: [{'thoughtWordCount': 30, 'actionWordCount': 5, 'verbosityRatio': 6.0, 'isZeroShot': false}]

{"highVerbositySteps": [{'thoughtWordCount': 30, 'actionWordCount': 5, 'verbosityRatio': 6.0, 'isZeroShot': false}], "count": 1}

A zero-shot indicator is triggered when the `analyze_trace_verbosity` tool finds that the thought block contains fewer than 10 words, suggesting the model skipped detailed reasoning.

Related Connectors