
MCP Peek
Open up your MCP servers and see what's really going on inside.
Demo Video
About MCP Peek
Native desktop app for macOS, Windows & Linux to inspect, test, and live-proxy Model Context Protocol (MCP) servers — with per-tool token context tracking.
MCP Peek is a native desktop app for macOS, Windows, and Linux that makes Model Context Protocol (MCP) servers visible. If you're wiring MCP servers into Claude Code, Cursor, or your own AI agents, you're normally working blind — config files, JSON-RPC over stdio, and cryptic connection errors. MCP Peek turns all of that into something you can see, click, and test before an agent ever touches it.
Connect to any MCP server over HTTP or stdio (npx, uvx, python, deno — any command) and MCP Peek lays out the server's full surface in one place: tools, resources, resource templates, and prompts. No more guessing what a server can actually do, or reading its source code just to find out what it exposes.
Running tools is where it saves the most time. Every tool gets a schema-aware form generated from its input schema, so you never hand-write request JSON. A skeleton generator gives you a valid starting payload in one click, and saved inputs let you re-run the same calls while you iterate on a server. Resource templates get a proper RFC 6570 inspector, so you can expand variables and fetch templated resources interactively.
The built-in proxy is the standout feature. Put MCP Peek between your MCP client and a server and watch the raw JSON-RPC flow stream in real time — every request, response, and notification, as it happens. The proxy also tracks cumulative context: the total tokens captured for the selected scope (a single server, or all servers combined), broken down per tool with repeated calls summed. That means you can see exactly which tools are quietly eating your agent's context window, and by how much, before your agent hits the wall mid-task.
When you've found what matters, flag the tools, resources, and prompts you care about and export them as a clean Markdown brief you can paste straight into your coding agent's context. A live token gauge shows what the brief will cost against your context budget before you export it.
MCP Peek is built with Tauri, so it's a real native app — small, fast, and lightweight on all three platforms, with code signing, notarization on macOS, and built-in auto-update. Privacy comes first: all network traffic runs in the native core (nothing routes through a web view), crash reports are scrubbed of sensitive data, and analytics are strictly opt-in.
Whether you're building your own MCP server and need a fast test loop, evaluating third-party servers before trusting them with your agent, or debugging why a server misbehaves inside a client, MCP Peek gives you the visibility that config files and log output never will.
Free to try. Pro is a one-time $19 purchase — no subscription — and unlocks multiple servers, the live proxy, unlimited saved inputs, and brief export.
Comments (1)
Similar products
AppLaunchFlow
Same categoryStore-ready app assets in minutes — screenshots, videos, ASO & more.

Status Monkey
Same categoryStatusMonkey monitors your websites for outages & alerts you

AI SaaS Directory Listing Generator
Same categoryGenerate structured SaaS copy from URL.
BuggyRun
Same categoryFind your app's vulnerabilities before attackers do.

ipynb.app
1 shared tagFree Online IPYNB (Jupyter Notebook) Converter

Aviation API
Same categoryThe Scalable Airline Logo API and Aviation Data Infrastructure for Modern Aviation Apps




@dworf · 7/31/2026
Hi everyone! I'm the maker of MCP Peek. I built it because I kept wiring MCP servers into my coding agents with no idea what they actually exposed — or why they silently failed. I wanted to click through a server's tools, run them without hand-writing JSON, and watch the live JSON-RPC flow, including which tools eat the most context tokens. It's a native app on macOS, Windows, and Linux. Happy to answer anything about MCP, Tauri, or building it — and I'd love your feedback!