Your agents use Argos end to end
Deploy a preview, read the diff, review the build, fix the flake. Every pillar is reachable from the CLI, the MCP server and the REST API, with the permissions of the person running the agent.
One agent, four jobs
Argos does four things for a pull request, and an agent can take each of them. Every pillar page has a section for agents with the full set of commands; these are the four it reaches for first.
Diff
Visual regression testing for screenshots, and diffs for any file: Markdown, JSON, and more.
The agent reads every diff as data: status, score, diff mask, baseline and current file.
argos build snapshots <build> --jsonReview
One place for humans and agents to approve what changed.
The agent approves, rejects or comments, attributed to the person whose token it runs with.
argos review create <build> --event approveStabilize
Kill flakes and debug failures with full per-test history.
The agent reads a test's flakiness and its recurring changes, then fixes the cause or ignores the change.
argos test changes <testId> --jsonDeploy
Free preview URLs for your Storybook or static site on every PR.
The agent ships its own preview with the command your CI runs, and the URL lands on the pull request.
argos deploy ./storybook-staticConnect your agent with the Argos MCP server
One remote server at mcp.argos-ci.com, streamable HTTP, nothing to install. Add the URL to your client, sign in with OAuth, and the agent lists builds, reads diffs, reviews changes and manages projects with the scopes you granted.
Scoped by OAuth
Your client opens a browser; you choose the organizations to share and the scopes to grant: projects:read, reviews:write, comments:write, projects:write, media:read, media:write, account:admin. Revoke it any time under Authorized applications.
Keep a human on writes
Approving a build or posting a comment goes through your client's approval prompt. Keep it on for write actions, and grant read-only scopes when the agent only inspects builds.
Learn moreTools generated from the REST API
Every API operation a user can call is a tool with the same name, parameters and permissions. When the management API shipped, the server went from 37 to 71 tools with no client update.
Learn moreTeach your agent the CLI in one command
The CLI runs wherever the agent has a terminal: your machine, CI, a sandbox. The skills teach it the commands, the flags, the token rules and a complete pull-request review, so you don't write the prompt.
argos-cli
The CLI's commands, flags, authentication rules and output formats, flakiness commands included, so the agent can investigate an unstable test and silence a change it can't fix.
Learn moreargos-pr-review
A complete review: find the Argos build from the pull request, inspect the snapshots that need review, summarize what changed, compare it with the intent, then approve or request changes.
Learn moreargos-upload
When a screenshot or a recording beats a paragraph, how to stage it on the branch so Argos posts it on the pull request, and how to read back the comments a human pinned on it.
Learn moreMachine-readable from the first request
The REST API describes itself with OpenAPI, this site answers in Markdown when asked for it, and the documentation ships every page as .md. An agent never has to scrape HTML to learn what Argos does or how to call it.
# One base URL, one bearer token
API=https://api.argos-ci.com/v2
AUTH="Authorization: Bearer $ARGOS_TOKEN"
# A test's changes; a project token can read
curl -H "$AUTH" \
$API/projects/acme/app/tests/<testId>/changes
# The whole API, described in OpenAPI 3.1
curl $API/openapi.yamlA REST API, described in OpenAPI
Everything the app does is at api.argos-ci.com/v2: builds and their diffs, reviews, comments, tests and their changes, deployments, projects, teams. One bearer token: a project token reads, a personal access token acts as a user. The OpenAPI 3.1 file is at /v2/openapi.yaml.
# Any product page as Markdown
curl https://argos-ci.com/md/review
curl -H "Accept: text/markdown" \
https://argos-ci.com/stabilize
# The site map, and how agents authenticate
curl https://argos-ci.com/llms.txt
curl https://argos-ci.com/auth.md
# Every docs page, and their index
curl https://argos-ci.com/docs/agents/mcp-server.md
curl https://argos-ci.com/docs/llms.txtThis site and the docs, in Markdown
Send Accept: text/markdown to a product page, or prefix its path with /md, and you get its Markdown twin. llms.txt maps the site, auth.md explains how agents authenticate, and the documentation publishes every page as .md, indexed by /docs/llms.txt.
OpenAPI and an API catalog
The homepage's Link header and /.well-known/api-catalog (RFC 9727) point at the OpenAPI description and the API docs, so a crawler finds the API without reading a page.
Markdown twins on every pillar page
/deploy, /diff, /review, /stabilize and this page each have one, with the page's FAQ appended. The docs add a read-only MCP server at argos-ci.com/docs/~gitbook/mcp.
WebMCP tools in the browser
In a browser with WebMCP, this site registers 12 read-only tools on document.modelContext: get_argos_overview, get_pricing, estimate_monthly_cost, get_page, search_docs, get_docs_page, get_quickstart, compare_with, get_security_compliance, get_latest_changelog, get_service_status, get_mcp_connection_info.
“Our agents open pull requests all day. Argos shows us exactly what each one changed, so reviewing AI-written code went from stressful to routine.”
Frequently asked questions
Which AI clients and agents work with Argos?
Any MCP client can connect to the Argos MCP server at mcp.argos-ci.com. The docs walk through Claude Code, Claude.ai and Claude for desktop, Cursor, VS Code with Copilot, Codex CLI, and Windsurf. Any agent with a terminal can use the CLI, locally or in CI, and the agent skills install into assistants that support skills, such as Claude Code, Codex, and Cursor.
What can an agent do over MCP, and what limits it?
The MCP server generates its tools from the REST API: every operation a user can call is a tool with the same name, parameters, and permissions. An agent can list builds and inspect diffs, approve or reject a build, request reviewers, read and post comments, investigate flaky tests and ignore their changes, configure projects and automations, administer a team, upload media, and read analytics and usage.
What it may actually do is decided by the OAuth scopes you grant, such as projects:read, reviews:write, comments:write, or account:admin, and by your client's approval prompt on write actions. See the available tools.
Which token does my agent need?
Uploads, deployments, and read-only commands such as build get, build snapshots, test list, test get, and test changes work with a project token, the one your CI already has. Anything attributed to a user, such as submitting a review, requesting reviewers, posting a comment, ignoring a change, or configuring a project, needs a personal access token, because the action is checked against that user's permissions. Locally, argos login stores one for you.
The MCP server takes OAuth or a personal access token and does not accept project tokens. See project tokens and personal access tokens.
Does Argos use AI to decide whether a change is acceptable?
No. Argos compares screenshots with deterministic pixel diffing, built on the open-source odiff library, and compares non-image files as text. Same input, same result, and the exact pixels that changed are visible. Agents use that build data as review evidence; they do not replace your team's ownership of review decisions. See how Argos detects visual differences.
Can an agent read this site and the documentation?
Yes. Send Accept: text/markdown to a product page, or prefix its path with /md, and you get its Markdown twin. llms.txt maps the site and auth.md explains how agents authenticate. Every documentation page exists as .md, indexed by /docs/llms.txt and /docs/llms-full.txt, and the docs expose a read-only MCP server at argos-ci.com/docs/~gitbook/mcp. In a browser with WebMCP, this site also registers read-only tools such as search_docs and get_pricing. See AI-ready docs.
Supercharge your product quality
See every change your team and your agents make. Review with confidence, and merge faster.