Claude Conversation Drift Detector

Claude Conversation Drift Detector MCP Connector for Claude

A+

Monitors AI agent focus by detecting task drift and topic shifts.

3 tools Official Updated Oct 1, 2026 Official Vinkius Partner

This MCP server provides tools to ensure AI agents stay aligned with their primary objectives. It identifies when an agent deviates from the user's intent by analyzing the semantic relevance of recent actions against the original task description. Using analyze_conversation_drift, agents can check if their current trajectory is on track. The detect_topic_shift tool identifies sudden jumps to unrelated subjects, while get_task_relevance_summary provides a percentage of task coverage based on keyword overlap. This is essential for preventing hallucination loops and tangential sub-tasks in complex workflows.

drift-detectionagent-monitoringsemantic-analysistask-alignmentai-reliability

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

get_task_relevance_summary

Provides a high-level summary of how much of the task has been addressed based on keyword coverage

analyze_conversation_drift

Determines if the current sequence of agent actions is staying aligned with the original user goal

detect_topic_shift

Identifies if the agent has abruptly switched to a completely unrelated topic

See how to talk to your AI agent using Claude Conversation Drift Detector.

Check if my current conversation is still focused on the original task.

The agent is currently on track with a relevance score of 0.85.

Has the agent switched to a new topic recently?

No sudden topic shifts have been detected in the recent action history.

How much of the task have I completed so far?

The task coverage is currently at 75%, with the following unaddressed keywords: [keyword1, keyword2].

It uses Jaccard similarity to compare keywords extracted from the original task with keywords extracted from recent agent actions like `analyze_conversation_drift` or file changes.

Related Connectors