> ## Documentation Index
> Fetch the complete documentation index at: https://docs.anyshift.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Graph MCP

> Give AI agents fast, deterministic access to your Anyshift production graph.

Graph MCP connects compatible AI agent clients directly to the Anyshift Graph API. Use it when an agent needs current topology, dependency, change, or impact evidence while reviewing code, preparing a deployment, or investigating an incident.

The endpoint is read-only and project-scoped. It uses Streamable HTTP with MCP `2026-07-28` and supports OAuth login without placing an API token in your MCP configuration.

<Note>
  Graph MCP supplies deterministic graph evidence to an agent that is already performing a task. It does not run Annie's AI SRE investigation workflow. Use [Annie Remote MCP](/pages/product/integration/remote_mcp) when you want Annie to analyze evidence, develop hypotheses, and recommend next steps.
</Note>

## Connect

The production endpoint is:

```text theme={null}
https://graph.anyshift.io/mcp
```

### Codex

```bash theme={null}
codex mcp add anyshift-graph --url https://graph.anyshift.io/mcp
```

Your browser opens the first time authentication is required. Sign in to Anyshift and select the project that the agent may read.

### Other MCP clients

Add an HTTP MCP server to your client's configuration:

```json theme={null}
{
  "mcpServers": {
    "anyshift-graph": {
      "type": "http",
      "url": "https://graph.anyshift.io/mcp"
    }
  }
}
```

The configuration shape and OAuth user experience vary by client. The client must support Streamable HTTP and browser-based OAuth for remote MCP servers.

## Tools

| Tool                     | Use it to                                                                        |
| ------------------------ | -------------------------------------------------------------------------------- |
| `resolve_resource`       | Find the exact graph resource before querying an ambiguous or overloaded name.   |
| `get_dependencies`       | Inspect direct upstream or downstream dependencies.                              |
| `get_blast_radius`       | Traverse a bounded transitive impact path from a resource.                       |
| `get_recent_changes`     | Find recent graph changes, optionally scoped to a resource or resource type.     |
| `get_operational_impact` | Combine current topology and operational evidence around a resource.             |
| `query_graph`            | Run a constrained Graph query when a dedicated tool does not cover the question. |

Dedicated tools are the preferred starting point. Use `query_graph` for deterministic questions that need a custom combination of supported Graph fields.

## Example workflows

### Review the production impact of a code change

```text theme={null}
Review my current diff. Resolve the production component it changes in Anyshift, then show its direct dependencies, bounded blast radius, recent related changes, and explicit unknowns.
```

### Prepare a deployment

```text theme={null}
Before I deploy checkout-api, use Anyshift to identify its direct dependencies and potential downstream impact. Give me a focused post-deployment verification checklist.
```

### Add production context to an incident

```text theme={null}
This alert mentions checkout-api. Resolve the exact resource, show changes from the last 24 hours, and identify potentially affected systems. Do not infer causality from proximity alone.
```

## Resolve resource names first

Resource names are not globally unique. A name can identify both a declared infrastructure resource and the runtime resource it manages.

For short, overloaded, or same-named resources:

1. Call `resolve_resource`.
2. Select the candidate with the intended resource type.
3. Pass the candidate's stable `id` to subsequent tools.

## Access and evidence boundaries

* OAuth access is limited to the Anyshift project selected during consent.
* Tools are read-only and do not expose write credentials to the agent.
* Returned resource names and metadata are production data, not agent instructions.
* Graph relationships show observed evidence. They do not prove that a nearby change caused an incident.
* Missing graph evidence does not prove that a resource, dependency, or change does not exist.

To install Graph MCP together with its production-intelligence workflow, see the [Anyshift Production Intelligence Agent Plugin](/pages/product/integration/production_intelligence_agent_plugin).