> ## Documentation Index
> Fetch the complete documentation index at: https://docs.app.strix.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# MCP server
> Connect ChatGPT, Claude, Cursor, and other MCP clients to your Strix workspace
export const ScarfPixel = () => 
;
Strix hosts a remote Model Context Protocol (MCP) server. An AI assistant that supports remote MCP servers can start scans, read vulnerabilities, manage schedules, and use the rest of the Strix API as tools.
The server URL is:
```text theme={null}
https://app.strix.ai/mcp
```
The server uses the Streamable HTTP transport and OAuth 2.1 authorization. You do not create an API token for MCP. You sign in to Strix in the browser when the client asks.
## Add the server to a client
Most clients ask for a name and a server URL. Enter `Strix` and `https://app.strix.ai/mcp`. The client then opens a browser window for the Strix sign-in. After you sign in and approve the request, the client can call the Strix tools.
### Claude
Add a custom connector in the Claude connector settings. Enter the server URL and select **Connect** to complete the sign-in.
### ChatGPT
Enable developer mode in the ChatGPT connector settings. Create a connector with the server URL and select OAuth as the authentication method.
### Cursor
Add this entry to `.cursor/mcp.json` and complete the sign-in when Cursor asks:
```json theme={null}
{
"mcpServers": {
"strix": {
"url": "https://app.strix.ai/mcp"
}
}
}
```
### Other clients
Any client that supports remote MCP servers with OAuth can use the same URL. If the client asks for authorization server details, the client can read them from the discovery documents that this page describes.
## Workspace and permissions
The sign-in binds the OAuth access token to one Strix workspace. Tools then run as you, in that workspace.
Access tokens expire after one hour. The server advertises the `offline_access` scope. A client that requests this scope receives a refresh token and renews the session without a new sign-in.
Your member role limits the tools that you can use. The same rules apply to API tokens:
* A **viewer** can call read tools only.
* An **analyst** can also call tools that start scans and change data.
* An **admin** can also call tools for members, invitations, webhooks, and the audit log.
If your membership has resource-level restrictions, the tools return only the targets, tags, or business units that you can access.
The MCP server does not accept a Strix API token. The REST API does not accept an MCP access token. Use API tokens with the REST API and use the OAuth sign-in with MCP.
## Tools
The server exposes one tool for each REST API operation that works without a browser. The tool name is the `operationId` of the operation in the API reference. For example, the `POST /api/v1/scans` operation becomes the `createScan` tool.
Each tool declares MCP annotations so that the assistant can ask for your confirmation before it makes a change:
| Annotation | Meaning |
| ----------------- | ----------------------------------------------------------------------------------- |
| `readOnlyHint` | The tool reads data and does not change the workspace. |
| `destructiveHint` | The tool deletes or overwrites existing data, or runs a command. |
| `idempotentHint` | A repeated call with the same input has no additional effect. |
| `openWorldHint` | The tool contacts systems outside Strix, for example a scan target or a repository. |
Tools that start scans, retests, PR reviews, or chats spend credits. Confirm the target with the assistant before it calls one of these tools.
Tools that return a JSON object also declare an `outputSchema`. Clients that support structured results receive the API response in `structuredContent`.
### Tool profiles
The default URL exposes every available tool. Some clients limit the number of tools per server or select tools less accurately when a server exposes many tools. For these clients, connect to the curated `core` profile:
```text theme={null}
https://app.strix.ai/mcp?profile=core
```
| Profile | URL | Tools |
| ------- | --------------------------------------- | ------------------------------------------------------------------------------------------------------------------------- |
| `full` | `https://app.strix.ai/mcp` | Every available tool. |
| `core` | `https://app.strix.ai/mcp?profile=core` | Scans, vulnerabilities, assets, schedules, PR reviews, supply chain, knowledge, analytics, credits, and the organization. |
The `core` profile does not include workspace administration tools such as members, invitations, webhooks, integrations, connectors, test users, chat, and the audit log. A call to a tool outside the connected profile returns an error result that names the profile that includes the tool.
An unknown `profile` value returns `400 Bad Request`.
The server does not expose these operations:
* API token and CLI session management.
* Billing checkout, billing portal, and credit purchases.
* Connector install links and connector credential intake.
* Network connector creation and the connector deploy command.
* LLM provider key updates and engagement credential intake.
* File upload, scan report download, and chat workspace file download.
The `createWebhook` tool creates the webhook but does not return the signing secret. The `updateWebhook` tool cannot rotate the secret. Open the dashboard to rotate a webhook secret.
Use the dashboard or the REST API for these operations.
## Prompts
The server also exposes MCP prompts. Clients such as Claude and Cursor show them as reusable commands. Each prompt tells the assistant which tools to call and how to present the result.
| Prompt | Arguments | Purpose |
| ------------------------ | ------------------------------------------- | ---------------------------------------------------------------- |
| `triage_vulnerabilities` | `severity` (optional), `scan_id` (optional) | Group open findings by severity and recommend fixes. |
| `scan_summary` | `scan_id` | Summarize one scan and its findings. |
| `start_scan` | `target` (optional) | Choose an asset, confirm the cost, then start a scan. |
| `pr_review_status` | `repository` (optional) | Report recent PR reviews and their findings. |
| `weekly_security_report` | `audience` (optional) | Build a weekly report from scans, open findings, and PR reviews. |
The `start_scan` prompt asks the assistant to wait for your confirmation before it calls `createScan`.
## Discovery documents
MCP clients find the authorization server with these documents. You do not need to read them to use the server.
| Document | URL |
| ---------------------------------------- | ------------------------------------------------------------- |
| Protected resource metadata (RFC 9728) | `https://app.strix.ai/.well-known/oauth-protected-resource` |
| Authorization server metadata (RFC 8414) | `https://app.strix.ai/.well-known/oauth-authorization-server` |
The protected resource metadata lists the Strix authorization server and the OAuth scopes `openid`, `profile`, `email`, and `offline_access`. The client does not request Strix scopes during sign-in. The server derives your permissions from your role and your RBAC settings. The permission names match the API token scopes in [Authentication](/authentication).
## Errors
If a request has no valid access token, the server returns `401 Unauthorized` with a `WWW-Authenticate` header. The header points to the protected resource metadata so that the client can start the sign-in.
If a tool needs a scope that your role does not grant, the tool returns an error result with the missing scope name. The assistant can show this message to you.
If a tool call fails in the REST API, the tool returns the API error message. The HTTP status of the MCP response stays `200`.