THE ADIOS MCP SERVER
Deploy from Claude, Cursor, or Codex.
You approve what ships.
Connect the AI you already use to Adios. It builds, previews, and checks your app in the cloud, and production deploys wait for your approval.
Works with Claude, Claude Code, ChatGPT, Cursor, VS Code, and more.
Source files
Builds & previews
Runtime logs
Reviewed releases
CLIENT DIRECTORY
Pick your AI tool.
Each guide shows the setup steps, the sign-in flow, and a first prompt to try.
FROM A PROMPT TO A RUNNING APP
See MCP in action.
Ask your AI client to build a page, investigate a problem, or publish an app. See how it can use Adios to get the work done.
Start in the AI tool you already use. Connect your Adios account, then describe what you want to do.
Find your clientBuild a landing page for my bakery and give me a preview link.
- create_workspace: Created a workspace for your bakery website.
- write_workspace_files: Added a home page, menu and contact section.
- prepare_workspace_preview: Build passed. The preview is ready.
Your first page is ready to explore. Example URL: https://bakery-preview.example
A FIRST TASK TO TRY
Ask for a preview.
Check the result.
Connect your client, authorize your account, then start with a bounded task.
COPYABLE STARTER PROMPT
Open my Adios workspace, inspect the source and build commands, and show me your plan before changing anything. After I approve it, build and check a preview. Do not deploy without my approval.
AVAILABLE TOOLS
From context to a checked release.
Expand a group to inspect the tools. Sensitive operations still require your confirmation.
Discover and authenticate / 2
Learn what the server can do and confirm the user and team behind the current OAuth session.
- adios_capabilities
- Describe the available Adios capabilities and tool groups.
- auth_me
- Return the current principal and team context.
Plan and track / 4
Keep the plan, progress, and important run events attached to the coding task instead of losing them between sessions.
- get_agent_state
- Read the session transcript, runs, plan steps, and events.
- save_agent_plan
- Persist the plan for the current coding task.
- update_agent_plan_step
- Mark a plan step passed, failed, blocked, or skipped.
- append_agent_event
- Record progress, tool, test, browser, approval, or deploy events.
Manage workspaces / 4
Create or open durable projects where the AI client can inspect source, make changes, and prepare a running preview.
- list_workspaces
- List workspaces available to the current team.
- get_workspace
- Read one workspace and its current state.
- create_workspace
- Create a workspace or a source-backed project shell.
- prepare_workspace_preview
- Write files, generate adios.yaml when needed, build, start, and wait for a preview URL.
Edit source / 10
Read and change project source while keeping the deployment contract explicit and destructive file operations approval-sensitive.
- list_workspace_files
- List files in a workspace.
- read_workspace_file
- Read one workspace file.
- write_workspace_file
- Write or replace one workspace file.
- write_workspace_files
- Write multiple workspace files together.
- write_workspace_adios_yaml
- Create or replace the deployment manifest.
- move_workspace_file
- Move or rename a file.
- delete_workspace_file
- Delete a file after explicit approval.
- stage_workspace_file
- Stage one workspace file for a commit.
- unstage_workspace_file
- Unstage one workspace file.
- check_workspace_adios_yaml
- Check manifest security and risky command patterns before execution.
Run and debug / 13
Run the same checks a developer would use, follow long-running command state, and diagnose the evidence from a failed build or preview.
- build_workspace
- Build an artifact from the workspace contents.
- build_workspace_dev
- Queue the canonical workspace build command.
- lint_workspace
- Queue the canonical lint command.
- start_workspace_run
- Start a preview runtime.
- stop_workspace_run
- Stop a preview runtime after approval.
- restart_workspace_run
- Restart the current preview runtime.
- test_workspace
- Run the declared unit tests.
- integration_test_workspace
- Run the declared integration tests.
- get_workspace_command_status
- Read command state by command ID.
- get_workspace_command_logs
- Read logs for a queued command.
- get_workspace_run_status
- Read the current preview runtime state.
- get_workspace_logs
- Read recent build or runtime logs.
- get_workspace_git_status
- Inspect current Git changes and source state.
Review and deploy / 4
Move a verified workspace into production through an explicit approval boundary, then inspect the live workload evidence.
- security_review_workspace
- Collect workspace context for a focused security review.
- deploy_workspace
- Deploy the current workspace after explicit approval.
- get_workload_logs
- Read recent logs for a deployed workload.
- deploy_workload
- Deploy an existing build after explicit approval.
Connected doesn’t mean unchecked.
Authorize the user and team. Keep source changes visible. Confirm sensitive actions.
CONNECTION DETAILS
Before you connect.
Endpoints and discovery
Remote HTTPhttps://api.adios.dev/v1/mcp
SSE fallbackhttps://api.adios.dev/v1/mcp/sse
Connection metadatahttps://api.adios.dev/v1/mcp/config
Tool discoveryhttps://api.adios.dev/v1/mcp/tools
Which MCP clients does Adios support?
Adios supports Claude, Claude Code, ChatGPT, Codex CLI, VS Code, Xcode, Cursor, Grok, Google Antigravity, GitHub Copilot CLI, Slack, OpenCode, Sublime Text, and clients that use the stdio bridge. Each has its own setup guide and FAQ.
Do I need an Adios account, and what does it cost?
Yes, a free Adios workspace. It needs no card and includes preview URLs; the Free preview sleeps after 4 hours idle. Always-on hosting with a custom domain starts at $10/month.
What MCP transport does Adios use?
The preferred public transport is Streamable HTTP at https://api.adios.dev/v1/mcp. Adios also publishes an SSE fallback and an npm-based stdio bridge for clients that launch a local command.
How is an Adios MCP session authenticated?
Adios uses OAuth. The session is scoped to the authorized user and team, and clients should complete the OAuth flow rather than storing a static platform token.
Can an MCP client deploy to production?
Yes, but deployment and other sensitive operations remain approval-gated. The client should stop, show what it intends to change, and wait for explicit confirmation before calling those tools.
Is this the same as hosting my own MCP server?
No. This page explains how AI clients use the hosted Adios platform through MCP. If you are building an MCP server of your own, use the Adios MCP server deployment guide.
YOUR NEXT DEPLOY STARTS HERE
Works on localhost? Ship it tonight.
Connect your AI, say “deploy this,” and share the link. Upgrade for production from $10/month.