Honest numbers
Every request shows where its time went, which certificate answered, where redirects went and the headers that really left your machine.
DNS8 ms
Connect17 ms
TLS27 ms
First byte72 ms
Download23 ms
Zorvik 0.2.0 is out · see what's newOne workbench
One workbench
for every wire.
Build, test, mock and load test APIs over HTTP, GraphQL, gRPC, WebSocket, Socket.IO, SSE, MQTT, TCP, UDP and DNS, test the MCP servers AI agents use, and see what really happens on the network.
Every protocol, one editor: params, auth, scripts and a response with timing, TLS and the headers that really left your machine.
- HTTP/1.1
- HTTP/2
- HTTP/3
- GraphQL
- GraphQL subscriptions
- gRPC
- MCP: Streamable HTTP & stdio
- WebSocket
- Socket.IO 3 & 4
- Server-Sent Events
- MQTT 3.1.1 & 5
- TCP
- UDP
- DNS over UDP, TCP, TLS, HTTPS
- OpenAPI 3
- Swagger 2
- Postman
- cURL
- OAuth 2.0
- mTLS
- JUnit
Everything in one place
One app instead of five.
A client for every protocol, tests, mocks, load tests and network tools, working from the same files and the same engine.
Tests that read like English
Postman-compatible scripts: imported collections run as they are.
pm.test("order is created", () => {
pm.response.to.have.status(201);
const order = pm.response.json();
pm.expect(order.id).to.be.a("string");
pm.environment.set("orderId", order.id);
});Mock anything
Routes with params and templates, delays, faults, CORS and forwarding.
- GET /users/:id 200
- POST /orders 201
- DELETE /orders/:id 503 · 1.5 s
- GET * → staging
Load tests on your laptop
Virtual users or arrival rate, thresholds like “p95 under 200 ms”.
Test MCP servers
See what an AI agent sees: an MCP server's tools, their schemas and answers. Call them over HTTP or stdio, test them in CI, or build an MCP server for your own agent to use.
✓ tools/call search_tickets structuredContent matches the schema
✓ tools/call refund_order isError: “older than 30 days”
✓ resources/read kb://refund-policy text/markdown
Your API work is plain files
A workspace is a folder of readable YAML. Commit it, review it in pull requests, share it. Secret values stay on your computer.
# requests/Orders/Create order.yaml
name: Create order
method: POST
url: {{baseUrl}}/orders
auth: { type: inherit }
+headers:
+ - { key: Idempotency-Key, value: {{$uuid}} }Built for AI agents
Claude Code, Codex, Gemini CLI, Cursor and others drive Zorvik over MCP while you watch. Risky actions wait for your OK.
Map the APIs in this repo and run them.
✓ save_requests 12 requests in “Orders API”
✓ run_collection 12 passed
? start_load_test waiting for your OK
Inspect the network
TLS inspector, DNS lookup, port check, ping, HTTP/3 check, and encoders for Base64, JWT, hashes and more.
Contract checks
Requests imported from an OpenAPI document are checked against it on every send. Update from a new version and keep your edits.
Works where you work
System or manual proxies, the OS certificate store, custom CAs and client certificates. Offline, with no account.
New · MCP testing
Test the tools your AI agents use.
AI apps and agents reach tools through MCP servers, and a model trusts whatever a tool's description, schema and answer say. Zorvik shows you exactly what the agent sees: connect to any MCP server, call its tools with your own arguments, watch every JSON-RPC message, and test the answers in CI. Building an agent? Give it an MCP server whose tools, failures and delays you control.
- Streamable HTTP, HTTP+SSE and stdio programs like
npx …, with auth and OAuth 2.0 - Tools with their schemas, resources, templates and prompts; tests and
zorvik run - MCP servers for AI apps, over HTTP or
zorvik serve … --stdio
MCP built in
Your coding agent drives it.
Connect Claude Code, Codex, Gemini CLI, Cursor or any MCP client with one command. The agent turns the APIs in your code into requests, sends them, writes tests, runs collections, builds and starts mock servers, runs load tests and reads the traffic, all inside Zorvik, where you watch every step.
- 38 tools: requests, environments, runs, servers, load tests, history, code export
- Deletes, load tests, servers and outside hosts wait for your OK
- Secrets are masked in everything the agent sees
$ claude mcp add --scope user zorvik -- zorvik mcpAdded stdio MCP server zorvik > Map the checkout API in this repo to Zorvik and test it✓ zorvik · save_requests 5 requests in "Checkout"✓ zorvik · run_collection 5 passed, 12 tests✓ zorvik · create_mock from openapi.yaml, port 4010? zorvik · run_load_test waiting for your OK in ZorvikFree, all of it
No paid plans. No account. No cloud.
Every feature is free for everyone, and it stays that way: no premium tier, no seat limits, no sign-in, and your work never leaves your computer unless you push it to your own Git.
Typical API clientsZorvik
FeaturesThe best ones behind a paid planEvery feature, free
AccountSign in to start, or to keep your workNone. Open it and work
Your collectionsSynced to the vendor's cloudPlain files in a folder you own
Sharing with your teamPaid team workspacesGit: pull requests, reviews, history
AI agentsTheir assistant, their cloudYour own agent over MCP, on your computer
What the app sends homeUsage analyticsNothing. Only the update check reads GitHub
New · Training Bootcamp
Learn it inside the app.
New to APIs and networks? A course takes you from “what is a network?” to mocking and load testing. Short readings with diagrams, then a lab in the real workbench: practice servers start on your computer and the Lab Guide ticks each step off the moment you get it right.
- Units
- 17
- Lessons
- 78
- Badges
- 25
- Hints from a nudge to the exact clicks, and “Do it for me”
- XP, levels, streaks and a Zorvik Bootcamp Graduate certificate
- Runs offline: every lab server lives on 127.0.0.1
Same engine, in your pipeline
The command line comes with the app.
Run collections in CI with JUnit reports, gate releases on load-test thresholds, serve mocks next to your tests, and connect AI agents.
$ zorvik run ./shop --env Staging --folder Checkout --junit report.xml✓ POST Create order https://staging.shop.example/orders 201 84 ms, 312 B ✓ status is 201 ✓ Matches the API spec (POST /orders → 201)✓ GET Get order https://staging.shop.example/orders/ord_81 200 12 ms, 488 B✗ POST Cancel order https://staging.shop.example/orders/ord_81/cancel 409 15 ms, 96 B ✗ can be cancelled — expected 409 to equal 200 Requests 2 passed, 1 failed, 0 skipped (3 total)Tests 3 passed, 1 failedTime 0.4 s$ echo $?1zorvik runCollections and folders, data files, JUnit and JSON reports. Exit code 1 when a test fails.zorvik loadA saved load test with its thresholds, and an HTML report for the build.zorvik serveAny saved mock or server, headless, with its traffic as JSON lines.zorvik mcpThe MCP server your AI agent starts to work with Zorvik.
Local first
No account, no cloud sync, no tracking. It works offline; the only connection it makes by itself is the update check to GitHub.
Files, not a database
Plain YAML in a folder you own. Git is your sync, history and review.
Secrets stay home
Secret values, cookies and tokens live on your computer, never in workspace files.
Open source
MIT or Apache-2.0, built in the open by the LibreGuild community.
Download
Get Zorvik.
Version 0.2.0 · released September 28, 2026 · what's new
Every download includes the app and the zorvik command line.
Windows
Your systemmacOS
Your systemLinux
Your systemOnly the command line
For CI machines, servers and containers: zorvik on its own, one file to unpack.
# In a pipeline: always the newest version
curl -fsSL https://github.com/LibreGuild/zorvik/releases/latest/download/zorvik-cli-linux-x86_64.tar.gz \
| sudo tar -xz -C /usr/local/bin zorvikThe app keeps itself up to date: it reads the newest version from this project's releases on GitHub, and sends nothing about you. Builds aren't code-signed yet: on Windows choose More info → Run anyway; on macOS allow it in System Settings → Privacy & Security. Want the newest changes? Try the nightly build, or build from source.