MCP without the
complexity.
Test an MCP server in 60 seconds.
Apple silicon·104 MB·Free
Built for the MCP ecosystem
- Speaks MCP revision 2025-11-25
- Reads .well-known/mcp.json (Server Cards)
- Runs locally
- Code-signed for macOS
- Free · No account
Press ⌘K
One window. Every server. Zero setup.
Open MCPOrbit, search the server you heard about, hit return. The protocol handshake, tool discovery, and parameter forms are all there waiting.
mcporbit — connect
Search MCP servers
⌘K- Open↵
Linear
Official PostgreSQL
OfficialSlack
CommunityNotion
CommunityFilesystem
Official
5 results
Download for macOS — free, no account, no telemetry.
create_issue
Create a new issue in a Linear team
"team": "ENG", "title": "Add dark mode", "description": "Users want..."
"id": 42, "url": "https://...", "state": "open"
Capabilities
Three jobs in one window.
Discovery, evaluation, and integration each have their own workflow today. MCPOrbit collapses them into a single surface.
Discover
Find every MCP server that exists
Browse the Official MCP Registry without leaving the app, and let MCPOrbit auto-discover servers from any domain that publishes a Server Card.
- Search the Official MCP Registry from inside the app
- Auto-discover servers via .well-known/mcp.json (Server Cards)
- Pluggable registry adapters for additional sources
Evaluate
See what a server actually does
Connect, browse the tools, run them with typed inputs, read the full response. No glue code, no agent in the loop.
- Auto-generated forms from each tool's JSON Schema
- Side-by-side comparison across servers and models
- A log of every tool call you run by hand, with request, response and duration
Integrate
Ship it into your stack
Copy a tested server's config block for Claude Desktop, Cursor, or Windsurf, with the file path and merge steps beside it. Export schemas as OpenAPI for everywhere else.
- Add to IDE for Claude Desktop, Cursor, Windsurf
- Export any server's schema as OpenAPI
- Postman-style environment variables with {{variable}} substitution
Server Cards
Decentralised discoveryAuto-discover MCP servers from any domain.
MCPOrbit reads .well-known/mcp.json from any domain you point it at, and every MCP server the card lists appears — no manual registration, no centralised directory. Think DNS for MCP tools.
- 01
The server publishes — drop a JSON file at
https://yourdomain.com/.well-known/mcp.json - 02
MCPOrbit reads it — point the app at the domain; every server shows up automatically.
linear
claude-opus-4.7287ms> tools/call list_issues
{ "team": "growth", "status": "open" }
< response
{ "issues": [4 items], "cursor": "..." }
postgres
gpt-5-mini412ms> tools/call list_issues
{ "team": "growth", "status": "open" }
< response
{ "issues": [4 items], "cursor": "..." }
Comparison Mode
One prompt. Every server. Every model.
Run the same task across any combination of MCP servers and language models. See outputs, tool-call traces, latency, and failures side by side. The eval workflow you currently fake with five terminal tabs.
- Test how claude-opus-4.7 and gpt-5 handle the same MCP server
- Diff structured outputs across two builds of the same server
- Spot performance regressions before they land in production
Use cases: vetting a new server before integration · regression testing during development · benchmarking model providers.
Agent panel
Bring your own model. Watch every tool call it makes.
MCPOrbit's Agent panel runs against any model provider you have credentials for — Anthropic, OpenAI, Google Gemini, OpenRouter, Groq, Ollama, LM Studio, AWS Bedrock, or any OpenAI-compatible endpoint. Point it at a connected MCP server and it drives real tool calls, showing you each one inline with the arguments the model actually sent.
Every tool call, inline
Tick the tools a run may reach, then watch each call the model makes with the arguments it actually sent.
Local + remote
Plug in ollama for fully-offline runs, or hit any OpenAI-compatible inference endpoint behind your VPC.
Add to IDE
From bench to your editor's config.
Once you've vetted a server in MCPOrbit, pick your editor. MCPOrbit validates the connection, generates the exact mcpServers block for it — command, arguments and env — and copies it to your clipboard.
The config file path and the merge-and-restart steps for that editor sit right beside the snippet.
Registry browser
The Official MCP Registry, inside the app.
Search every active server in the Official MCP Registry without ever opening a browser. MCPOrbit caches results locally, supports lazy pagination, and adds new sources as the ecosystem grows.
mcporbit — registry
Search the Official MCP Registry
all serverslinear
Issues, projects, and cycles
postgres
Query, schema, migrations
github
PRs, issues, code search
slack
Channels, messages, files
notion
Pages, databases, blocks
filesystem
Read, write, watch local files
Built for builders
The smaller details that add up.
Every shipped feature that doesn’t need its own section. Each one quietly removes a step you used to do by hand.
Environment variables
Postman-style {{variable}} substitution across connections, headers, and tool inputs. Define dev, staging, prod once. Use them everywhere.
Request/response log
Every tool call MCPOrbit makes against a server, timestamped and inspectable — the arguments you sent, the full response JSON, and the duration.
Copy as curl
One-click export of any tool call as a curl command. Or as the raw schema. Or as the response.
OpenAPI export
Export any connected server’s schema as OpenAPI. For docs, for Postman collections, for any other tool that doesn’t speak MCP yet.
Schema explorer
Every tool’s JSON Schema rendered as a readable form, with required fields, defaults, enums, and inline descriptions surfaced.
Code-signed and notarized
No Gatekeeper warnings, no right-click-to-open. Apple silicon build, signed and notarized.
Auto-updates
Background updates via electron-updater. Always on the latest release, never reinstalling DMGs.
No telemetry, no analytics
MCPOrbit ships neither. It does check for updates automatically on launch, and the Privacy Policy sets out exactly what that request sends. The future opt-in telemetry that powers the Popular MCP Servers report has a public schema and a one-click delete.
Free, no account
Free to download and use, at home or at work. No sign-up, no paid tier, no seat count. MCPOrbit stores your servers and credentials on your own machine, and sends a credential only to the server or provider you gave it to.
Time-to-value
From curiosity to confidence.
Connect a server you heard about.
Paste the launch command or URL. MCPOrbit handles the handshake, capability negotiation, and pulls in every tool. You see the server work in a couple of seconds.
Vet it before it hits production.
Run the tools manually. Compare two models on the same task. Watch which tools each one picks. Decide whether to integrate based on what the server actually does, not what the README claims.
Ship it into your stack.
Copy the generated config block for Claude Desktop, Cursor, or Windsurf and merge it into your editor’s config. Or export the schema as OpenAPI. By the time you’ve integrated, you’ve already seen every tool fail in every way it can.
Honestly
How does MCPOrbit fit in?
| Approach | Speed | Trust |
|---|---|---|
| Clone the README + write a client | Slow | Hope-driven |
| Use Postman with a custom collection | Medium | Doesn’t model MCP primitives |
| Browse a registry directory in a browser | Fast to find | Can’t actually run anything |
| Open MCPOrbit and connect | Seconds | Run, compare, ship |
We're not trying to replace Postman. We're trying to do what Postman did for HTTP — but for MCP.