Submit

MCP Peek

@Comis

MCP Peek is a local-first desktop app to inspect, run, and proxy MCP servers. Connect any server over stdio or HTTP/SSE, browse its full catalog of tools, resources, and prompts with token-cost estimates, run tools through auto-generated forms, and capture live request/response traffic with the built-in proxy.
Overview
     (no-JS users see [data-reveal] content fully visible). --><script>document.documentElement.classList.add('js')</script><meta charset="utf-8"><!-- Preload critical font (Geist Sans 400) --><link rel="preload" href="/_astro/geist-sans-latin-400-normal.gapTbOY8.woff2" as="font" type="font/woff2" crossorigin><!-- Favicons --><link rel="icon" href="/favicon.svg" type="image/svg+xml"><link rel="icon" href="/favicon.ico" sizes="32x32"><link rel="apple-touch-icon" href="/apple-touch-icon.png"><title>MCP Peek — Inspect, run &amp; proxy MCP servers</title><meta name="description" content="MCP Peek is a desktop app to inspect, run, and proxy Model Context Protocol (MCP) servers. See every tool, resource and prompt, run them, and capture traffic."><meta name="viewport" content="width=device-width, initial-scale=1"><meta name="theme-color" content="#0C1322"><link rel="canonical" href="https://mcppeek.com/"><!-- Open Graph --><meta property="og:title" content="MCP Peek — Inspect, run &#38; proxy MCP servers"><meta property="og:description" content="MCP Peek is a desktop app to inspect, run, and proxy Model Context Protocol (MCP) servers. See every tool, resource and prompt, run them, and capture traffic."><meta property="og:url" content="https://mcppeek.com/"><meta property="og:type" content="website"><meta property="og:image" content="https://mcppeek.com/og-default.png"><!-- Twitter --><meta name="twitter:card" content="summary_large_image"><meta name="twitter:title" content="MCP Peek — Inspect, run &#38; proxy MCP servers"><meta name="twitter:description" content="MCP Peek is a desktop app to inspect, run, and proxy Model Context Protocol (MCP) servers. See every tool, resource and prompt, run them, and capture traffic."><meta name="twitter:image" content="https://mcppeek.com/og-default.png"><link rel="preload" href="/_astro/app-overview-dark.DtInuB_c_1VJcOs.webp" as="image" type="image/webp"><style>:root{--accent:#38BDF8;--glow:.7;--navy:#0C1322;--deep:#070C16;--panel-a:#0B1626;--panel-b:#0A1320;--indigo:#4B57D6;--link:#0369A1;--link-deep:#075985;--surface:#F6F8FB;--surface-alt:#EAF0F7;--surface-stage:#F7F9FC;--card:#fff;--border:#E6EBF1;--border-strong:#E3E8EF;--border-soft:#DEE6F0;--text:#161A21;--text-2:#1A1F27;--muted:#5A6473;--muted-2:#6B7480;--on-dark:#EAF1F8;--on-dark-hi:#F7FBFE;--st-explore:#0284C7;--st-run:#4B57D6;--st-read:#7C3AED;--st-budget:#0D9488;--st-brief:#D97706;--ok:#34D399;--warn:#F8A34A;--err:#F87171;--maxw:1200px;--font-sans:"Geist Sans","Geist",system-ui,-apple-system,sans-serif;--font-mono:"Geist Mono",ui-monospace,SFMono-Regular,monospace}*{box-sizing:border-box}html{scroll-behavior:smooth}::selection{background:#38bdf847}@media(prefers-reduced-motion:reduce){:root{--glow:.35}html{scroll-behavior:auto}*,*:before,*:after{animation:none!important;transition:none!important}}
Inspect & debug · Model Context Protocol

See inside your
MCP servers.

The desktop tool for inspecting and debugging MCP servers. Connect to any server, run its tools by hand, and read the real request, response, and timing behind every call — no model in the middle, no guessing.

$19 Pro — no subscription · macOS · Windows · Linux

mcp-peek · Useful (HTTP) · 153 tools · ~29.5K tok
MCP Peek main window — server catalog sidebar, the selected OMDb-get_by_imdb_id tool with its auto-generated schema form, and the rich JSON result panel.
stdio· HTTP + SSE· OAuth / bearer· tools · resources · templates · prompts· live token counts
A development tool

Talk to the server. Not a model's idea of it.

MCP Peek is a developer tool for inspecting and debugging MCP servers — not a hub, gateway, or multiplexer. Asking an assistant to "try the tool" shows you what the model did with it — paraphrased, lossy, one step removed. MCP Peek connects straight to the server, so when you're building or debugging you work with the real thing:

Debug — reproduce any call and read the raw request and response, not a summary.
Inspect — every tool, schema, resource and prompt, exactly as the server defines it.
Understand — see how a server actually responds before you wire it into an agent.
Proxy — sit between your agent and its servers and watch every call it makes, live.
Through an LLM
You AI Client MCP server
interpreted · lossy · you trust the paraphrase
You MCP Peek MCP server
You AI Client MCP Peek MCP server
The scope

Everything a server exposes — on one bench.

View as
Browse every tool, resource, template and prompt — each with a live token count.
Server catalog
Server catalog — Tools, Resources, Templates and Prompts tabs with counts, a filter field, and a list of tools each showing its token count.Add server dialog set to HTTP transport — server URL, authentication, custom and passthrough headers.Add server dialog set to stdio transport — command, arguments and environment variables.
Pro · The Proxy

Sit in the middle. Watch every call.

Point Claude or Cursor at MCP Peek while you build, and watch the signal move — every call, its timing, tokens, and full request/response, plus a topology view and per-flow detail. A local debugging passthrough, not a production gateway.

Live flow capturing serv-everything
Filter flow — method, server, status…
TimeRoute · methodStatusmstokens
16:01:30 tools/call trigger-long-running-operation BoltAI → serv-everything 200 1013 74
16:01:30 tools/call get-sum BoltAI → serv-everything 200 1 77
16:01:29 tools/call get-resource-links BoltAI → serv-everything 200 1 225
16:01:28 tools/call toggle-subscriber-updates BoltAI → serv-everything 200 1 51
16:01:27 tools/call toggle-simulated-logging BoltAI → serv-everything 200 1 51
16:01:23 tools/call toggle-subscriber-updates BoltAI → serv-everything 408 30021 0
16:01:23 tools/call simulate-research-query BoltAI → serv-everything 200 1 88
16:01:22 tools/call gzip-file-as-resource BoltAI → serv-everything 200 241 107
16:01:21 tools/call simulate-research-query BoltAI → serv-everything 200 0 91
16:01:20 tools/call gzip-file-as-resource BoltAI → serv-everything 200 25 195
16:01:14 tools/call trigger-long-running-operation BoltAI → serv-everything 200 3007 74
16:01:11 tools/call trigger-elicitation-request BoltAI → serv-everything 500 4 63
16:01:11 tools/call trigger-sampling-request BoltAI → serv-everything 200 1 87
16:01:09 tools/call gzip-file-as-resource BoltAI → serv-everything 200 1 156
16:01:02 tools/call get-resource-links BoltAI → serv-everything 200 3 633
16:01:02 tools/call get-structured-content BoltAI → serv-everything 422 2 41
16:01:01 tools/call get-structured-content BoltAI → serv-everything 200 1 97
16:00:54 tools/call get-annotated-message BoltAI → serv-everything 200 2 3.7K
16:00:49 tools/call get-tiny-image BoltAI → serv-everything 200 3 3.7K
16:00:48 tools/call get-env BoltAI → serv-everything 200 3 2.1K
16:00:37 tools/list BoltAI → serv-everything 200 6 2.6K
16:01:30 tools/call trigger-long-running-operation BoltAI → serv-everything 200 1013 74
16:01:30 tools/call get-sum BoltAI → serv-everything 200 1 77
16:01:29 tools/call get-resource-links BoltAI → serv-everything 200 1 225
16:01:28 tools/call toggle-subscriber-updates BoltAI → serv-everything 200 1 51
16:01:27 tools/call toggle-simulated-logging BoltAI → serv-everything 200 1 51
16:01:23 tools/call toggle-subscriber-updates BoltAI → serv-everything 408 30021 0
16:01:23 tools/call simulate-research-query BoltAI → serv-everything 200 1 88
16:01:22 tools/call gzip-file-as-resource BoltAI → serv-everything 200 241 107
16:01:21 tools/call simulate-research-query BoltAI → serv-everything 200 0 91
16:01:20 tools/call gzip-file-as-resource BoltAI → serv-everything 200 25 195
16:01:14 tools/call trigger-long-running-operation BoltAI → serv-everything 200 3007 74
16:01:11 tools/call trigger-elicitation-request BoltAI → serv-everything 500 4 63
16:01:11 tools/call trigger-sampling-request BoltAI → serv-everything 200 1 87
16:01:09 tools/call gzip-file-as-resource BoltAI → serv-everything 200 1 156
16:01:02 tools/call get-resource-links BoltAI → serv-everything 200 3 633
16:01:02 tools/call get-structured-content BoltAI → serv-everything 422 2 41
16:01:01 tools/call get-structured-content BoltAI → serv-everything 200 1 97
16:00:54 tools/call get-annotated-message BoltAI → serv-everything 200 2 3.7K
16:00:49 tools/call get-tiny-image BoltAI → serv-everything 200 3 3.7K
16:00:48 tools/call get-env BoltAI → serv-everything 200 3 2.1K
16:00:37 tools/list BoltAI → serv-everything 200 6 2.6K
Topology 1 endpoint · 5 servers
LLM clients
1 connected · BoltAI
MCP over HTTP
request
MCP Peek proxy
127.0.0.1:8787
open · no auth
inspects & logs
observed
All servers /mcp/
serv-everything stdio
BrightData HTTP
Useful HTTP
Hindsight HTTP
instantdomainsearch.com HTTP
tools/call get-sum 200 BoltAI → proxy → serv-everything
Find in request & response…
Round trip 1 ms
proxy in 0 ms server processing 1 ms return 0 ms
Request / Response
nested
REQUEST
{
  "arguments": {
    "a": -15.5,
    "b": 0.5
  },
  "name": "get-sum"
}
RESPONSE
{
  "content": [
    {
      "text": "The sum of -15.5 and 0.5 is -15.",
      "type": "text"
    }
  ]
}

alt: proxy view — live flow log of MCP calls with timing & tokens, beside a topology graph of the servers it sits in front of and inspects.

Local-first · private

Your signal never leaves the machine.

Connections, secrets, and results stay local. Keys live in your OS keychain. The app does its work on your machine — analytics are off by default, and the UI makes no network calls of its own.

Keys in your keychain
Credentials handed to the OS, never to us.
No telemetry by default
Analytics off until you opt in.
Results stay on disk
Every call and response is yours alone.
Pricing

Pay once. Yours forever.

Not a subscription. One license, all future updates.

Free
$0
Explore + run, forever.
Download — free
1 server
3 saved inputs
Full explore — catalog of tools, resources, prompts
Run tools with auto-generated forms
Rich result views + find-in-result
Pay once
Pro
$19 once · no subscription
Everything, unlocked.
Get Pro — $19 once
Unlimited servers
Unlimited saved inputs
Export Markdown briefs
The Proxy — inspect live traffic + flow log
Topology + per-flow detail
All future updates
FAQ

Frequently asked questions

Everything you might want to know before you download.

What is MCP Peek?

A desktop app for inspecting, exercising, and proxying MCP (Model Context Protocol) servers. Connect any server, browse its tools, resources, templates and prompts with token-cost estimates, run them, capture live traffic, and export review briefs for coding agents.

What is MCP (Model Context Protocol)?

An open standard that lets AI clients connect to external tools and data through “servers.” MCP Peek is a GUI for exploring and debugging those servers.

Which platforms are supported?

macOS (Apple Silicon and Intel), Windows, and Linux.

What’s included for free?

The full inspector on one server — browse the catalog, run tools, and flag items. Everything except the Pro features.

What do I get with Pro?

Unlimited servers, the proxy (live traffic capture), review-brief export, and unlimited saved input presets.

How much is Pro — is it a subscription?

$19, one time. No subscription. One license, all future updates — pay once, it’s yours forever.

What kind of MCP servers does it work with?

Any of them: local processes over STDIO, or remote HTTP endpoints, with bearer-token, OAuth, or custom-header authentication.

What is the proxy / “Live flow”?

A local endpoint you point your AI client at. MCP Peek sits between the client and your servers and captures every request and response, so you can see exactly what’s being sent.

Is my data private? Does it phone home?

MCP Peek runs locally — your server configs and traffic stay on your machine, and the UI makes no network calls of its own. Crash reports (EU-hosted, stripped of arguments, URLs and secrets) are on by default and can be turned off in settings; product analytics are off by default and opt-in.

What’s a “review brief”?

Flag any tool, resource, or prompt that needs changing — optionally attaching its schema, input, and last response — and export a clean Markdown brief to hand to a coding agent.

Still have questions? Get in touch.

Download MCP Peek

Free to explore and run. $19 once unlocks unlimited servers and the Proxy.

macOS universal (Apple Silicon & Intel) · Windows 64-bit · Linux 64-bit

© 2025 MCP.so. All rights reserved.

Build with ShipAny.