Annie Remote MCP

Annie Remote MCP lets AI coding assistants run Annie investigations over HTTP with OAuth authentication. No local dependencies or API tokens are required. Your editor handles login automatically.

Use Annie Remote MCP when you want the AI SRE agent to investigate infrastructure, analyze logs, debug incidents, and produce recommendations. If your agent needs fast, deterministic production graph evidence while it performs another task, use Graph MCP instead.

Setup

One-command install

Install across all your editors at once using add-mcp:

npx add-mcp https://api.anyshift.io/mcp/remote -n annie -g

This auto-detects your installed editors (Claude Code, Cursor, Windsurf, VS Code, Claude Desktop, etc.) and configures them all.

Manual configuration

Add to your editor’s MCP config file (e.g. .cursor/mcp.json, ~/.claude.json, claude_desktop_config.json):

{
  "mcpServers": {
    "annie": {
      "type": "http",
      "url": "https://api.anyshift.io/mcp/remote"
    }
  }
}

First Use

On your first Annie tool call, your editor will open a browser window:

Log in with your Anyshift account
The connection is established automatically
No tokens to copy — your editor stores the credentials

That’s it. All subsequent tool calls are authenticated automatically.

Available Tools

Asking Annie

Tool	Description
`ask_annie`	Ask a question about your cloud infrastructure. Pass `mode="rca"` to run a multi-step root-cause investigation instead of a chat. Pass `hint="report"` to format the answer as report blocks, or `hint="diagram"` for a Mermaid diagram.
`get_annie_response`	Poll for Annie’s answer (called automatically by your editor)
`list_projects`	List your Anyshift projects
`set_project`	Switch the active project for queries

Custom Reports

After an ask_annie call with hint="report", you can save the answer as a reusable report definition and re-run it later.

Tool	Description
`save_report_definition`	Save a report-shaped answer as a named definition
`list_report_definitions`	List the report definitions in the active project
`list_report_instances`	List past runs of a specific definition
`get_report_instance`	Fetch a generated report as markdown or JSON
`generate_report`	Trigger a fresh run of a definition. The backend dedupes if a generation is already pending.

Feedback

Ratings flow back into Annie so future investigations can take your judgment into account.

Tool	Description
`submit_answer_feedback`	Thumbs up or down on a chat or RCA answer
`submit_hypothesis_feedback`	Thumbs up or down on a single hypothesis inside an RCA. The rating reaches the agent as a live update.

Pre-built skills for Claude Code

If your agent runs in Claude Code, install the Annie Skills plugin to skip the prompt-engineering work. The plugin ships ready-made skills that teach Claude Code when to call which MCP tool, how to chain calls, and how to recover from auth or staleness failures. Source: anyshift-io/annie-skills. This is separate from Production Intelligence for AI Agents, which packages Graph MCP and a portable production-evidence workflow for compatible agent clients.

Example Prompts

Use Annie to investigate the spike in 5xx errors on the payments service

Ask Annie what changed in our infrastructure in the last 24 hours

Ask Annie to trace the dependency chain for the prod RDS instance

Ask Annie why the Lambda function can't connect to the database

Annie typically takes 30 seconds to a few minutes depending on the complexity of the question. Your editor will poll automatically until the answer is ready.

Managing Access

You can view and revoke authorized MCP connections from your profile page under Authorized Apps.

Troubleshooting

Re-authenticate

If your session expires, reset the connection and try again:

# Claude Code
claude mcp reset annie

Your next Annie tool call will re-open the browser for login.

Browser doesn't open during auth

Make sure your default browser is set in your OS settings. If using a remote/SSH session, you may need to use the local stdio server instead.

Timeout waiting for response

This is normal — Annie is still working. The get_annie_response tool is called automatically by your editor until the answer is ready. Complex investigations can take a few minutes.

Get Started

Product

Manage

Setup

One-command install

Manual configuration

First Use

Available Tools

Asking Annie

Custom Reports

Feedback

Pre-built skills for Claude Code

Example Prompts

Managing Access

Troubleshooting

​Setup

​One-command install

​Manual configuration

​First Use

​Available Tools

​Asking Annie

​Custom Reports

​Feedback

​Pre-built skills for Claude Code

​Example Prompts

​Managing Access

​Troubleshooting

Setup

One-command install

Manual configuration

First Use

Available Tools

Asking Annie

Custom Reports

Feedback

Pre-built skills for Claude Code

Example Prompts

Managing Access

Troubleshooting