Stop guessing how LLMs use your MCP server.
No more scrolling through logs to reconstruct MCP usage.
Inspectr makes Model Context Protocol tool calls, prompts, resources, and tokens observable across Claude, OpenAI, ... with one command.
Understanding real MCP usage is harder than it looks
An MCP server can work functionally correct during development and still be difficult to understand once it’s used by real LLM clients.
In practice, teams rely on logs to understand what happened; but logs make it hard to reconstruct actual LLM behavior.
Questions like these are difficult to answer from logs alone:
- Which MCP calls are actually made by the LLM/Agents?
- What MCP tools, prompts, or resources were requested, and what output were returned?
- Which flow or sequence of tool calls did the model follow?
- How many tokens were consumed during a conversation?
The challenge is no longer if the MCP server works; but understanding what the LLM actually did.
Inspectr provides MCP insights
Flow
Inspectr runs as a transparent, pluggable proxy in front of your MCP server. It captures and understands MCP traffic in real time without requiring any code changes.
Full capture of MCP requests
See exactly what was sent and returned across every MCP operation.
MCP flow visibility
Follow the real execution flow, not fragmented log lines.
Call classification
Know whether tools, prompts, or resources were invoked.
Token usage estimates
Identify expensive flows and unexpected token spikes.
Exportable JSON sessions
Share, review, or investigate MCP sessions offline.
Guided MCP analysis
Surface patterns, anomalies, and behavior changes fast.
MCP insights in practice
MCP tool list
MCP tracing flow
MCP Tool call
Token usage
MCP request details
Get MCP visibility in seconds
Works with Claude, OpenAI, and other MCP clients. No SDKs, no instrumentation, and no server changes.
npx @inspectr/inspectr --backend=<https://your-mcp-server:3000>
Drop-in for local development or remote testing with a single command.
Build MCP servers with confidence
Understand real MCP usage before it reaches production and gains insights in LLM behavior.