Onboard your agent to Snapvisor

Fastest: Add to Claude

Using claude.ai, Claude Desktop or Claude Code signed in with a Claude account? Add Snapvisor as a connector in one click, then sign in to approve access.

Paste one sentence

The sentence points the agent at /agent-setup/prompt.md, a plain markdown file with the exact commands below. The agent handles setup and reports any sign-in, restart, or CI secret steps you need.

Guides per agent

How the credentials work

• MCP server, OAuth 2.1 (default). A request without a token gets a 401; compatible clients offer browser sign-in to Snapvisor. The client handles token storage after you approve access. The grant appears under Settings, Applications in the app and can be revoked there.
• MCP server, personal access token (headless alternative). Use it only when the user asks for a token-based setup or your client cannot open a browser. If SNAPVISOR_ACCESS_TOKEN is not already set, ask the user ONCE to create a token at https://app.snapvisor.io under Settings, Tokens, and to export it as SNAPVISOR_ACCESS_TOKEN in their shell profile. Never write the token into a file that is committed.
• Project tokens are rejected by the MCP server. A project's ARGOS_TOKEN only uploads screenshots from CI (step 3). It cannot read the API or drive the MCP server.
• Plan. Hosted MCP access through OAuth requires a paid plan; eligible new teams can start a 14-day Pro trial. On Free, getMe still verifies the connection; other OAuth operations require an upgrade. Personal access tokens use their account permissions separately.

Any other MCP client

1. Point the client at https://mcp.snapvisor.io over streamable HTTP. Clients that use an mcpServers file take this entry. A client that cannot run OAuth sends Authorization: Bearer followed by a personal access token.
   jsonCopy

   {
     "mcpServers": {
       "snapvisor": {
         "url": "https://mcp.snapvisor.io"
       }
     }
   }

Add the SDK

1. Only do this step inside a project that has a UI test suite; otherwise skip it and say so in the completion message. Use the package manager the project already uses (the lockfile tells you): the commands below are written for npm.
2. Playwright (@playwright/test in package.json): install the SDK, register its reporter in playwright.config.ts, and capture screenshots with argosScreenshot in the tests that render UI.
   bashCopy

   npm install --save-dev @snapvisor/playwright

3. The reporter uploads from CI only, so local runs stay offline:
   tsCopy

   // playwright.config.ts
   import { defineConfig } from "@playwright/test";
   import { createArgosReporterOptions } from "@snapvisor/playwright/reporter";
   
   export default defineConfig({
     reporter: [
       process.env.CI ? ["dot"] : ["list"],
       [
         "@snapvisor/playwright/reporter",
         createArgosReporterOptions({ uploadToArgos: !!process.env.CI }),
       ],
     ],
   });

4. Storybook: install @snapvisor/storybook. Vitest browser mode: install @snapvisor/vitest. Then wire them as the SDK page describes.
   bashCopy

   npm install --save-dev @snapvisor/storybook

5. Any other runner that writes screenshots to a folder: install the CLI and upload that folder after the tests run.
   bashCopy

   npm install --save-dev @snapvisor/cli
   npx @snapvisor/cli upload ./screenshots

6. Python projects: follow the Python SDK section of https://snapvisor.io/docs/sdks.
7. Uploads authenticate with the project's ARGOS_TOKEN, found in the project's settings at https://app.snapvisor.io. It belongs in CI secrets, never in the repository. On GitHub Actions no token is needed: https://snapvisor.io/docs/learn/integrations/github-tokenless-authentication.
8. Optional: the CLI can also inspect builds, submit reviews and post comments from the terminal. login opens the browser for the user to approve; whoami confirms it.
   bashCopy

   npx @snapvisor/cli login
   npx @snapvisor/cli whoami

Verify the connection

1. The snapvisor MCP server normally shows only two tools, search_tools and execute_typescript, and every operation is reached through them. Call search_tools with the query getMe; it answers with the declaration of external_getMe. Then call execute_typescript with this program:
   tsCopy

   return await external_getMe({});

2. If the server instead lists getMe as a tool of its own, call that tool directly. Either way, success is a response naming the signed-in user and the accounts the connection can reach; each account slug is the owner argument of every other operation.
3. Most agents load MCP servers when a session starts. If no snapvisor tool is available to you yet, do not report failure: the completion message tells the user to restart the session and ask you to verify the connection.
4. A 401 means the browser sign-in has not been completed yet. Inside execute_typescript a refused call throws an error carrying the same message; wrap the call in try/catch to read it. An answer that asks for an upgrade means the connection works and the account is on the Free plan.