MCP Server (Experimental)

fsPulse includes a built-in Model Context Protocol (MCP) server that allows AI agents to query scan data, analyze integrity issues, and explore filesystem history.

Experimental: MCP support in fsPulse is experimental. You may experience connectivity issues depending on your client and connection method. See Setup for details on known limitations.

The MCP endpoint is served at /mcp on the same port as the web UI. It uses the Streamable HTTP transport, compatible with Claude Desktop, Claude Code, and other MCP clients.

What Can an Agent Do?

The agent has access to fsPulse’s full data model — roots, scans, items, versions, and hashes — through 10 tools. It can:

• Explore — browse directory trees and search for files at any point in time
• Query — run structured queries with filtering, aggregation, and ordering across all domains
• Analyze — investigate integrity issues, track storage growth, and identify high-churn files
• Report — summarize activity over time periods, compare scans, and generate trend data

The most effective use is iterative: start with a broad question (“what changed this week?”), then drill into specific folders, files, or time ranges based on what the agent finds. See Sample Prompts for examples.

Pagination

All tools return at most 200 rows per call. The agent handles pagination automatically — if results are truncated, it can request the next page. You don’t need to think about pagination in your prompts; the agent manages it as needed.

Contents

• Setup — Enable MCP and configure your client
• Sample Prompts — Example prompts and multi-step investigation workflows
• Tools — The 10 tools available to AI agents