Computer Use File Server + MCP

Recommended: LiteLLM MCP Proxy

Endpoint: http://localhost:8081/mcp/docker_ai

Use your regular LiteLLM API key (NOT the docker-ai internal key).

Available Tools

bash_tool — execute bash commands in an isolated Docker container
view — view files and directories
create_file — create new files
str_replace — edit files (text replacement)
sub_agent — delegate tasks to an autonomous agent (Claude)

Usage Examples

Step 1: Initialize (obtain session ID)

curl -sD - -X POST "http://localhost:8081/mcp/docker_ai" \
  -H "Authorization: Bearer <LITELLM_API_KEY>" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-Chat-Id: my-unique-session" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}'

# Response contains header: mcp-session-id: <SESSION_ID>

Step 2: Call a tool

curl -s -X POST "http://localhost:8081/mcp/docker_ai" \
  -H "Authorization: Bearer <LITELLM_API_KEY>" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: <SESSION_ID>" \
  -H "X-Chat-Id: my-unique-session" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"bash_tool","arguments":{"command":"echo Hello","description":"test"}}}'

Required Headers

Header	Description
`Authorization`	Bearer token from LiteLLM: `Bearer <LITELLM_API_KEY>`
`Accept`	`application/json, text/event-stream`
`X-Chat-Id`	Unique session ID (creates a separate Docker container)
`Mcp-Session-Id`	Session ID from the initialize response
`X-MCP-Servers`	(optional) Comma-separated list of MCP servers for sub_agent (see below)

Usage via chat/completions

curl -X POST http://localhost:8081/v1/chat/completions \
  -H "Authorization: Bearer <LITELLM_API_KEY>" \
  -H "X-OpenWebUI-Chat-Id: my-chat-123" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude/claude-sonnet-4-5",
    "messages": [{"role": "user", "content": "Run ls -la"}],
    "tools": [{"type": "mcp", "server_label": "docker_ai"}]
  }'

MCP Servers for sub_agent

To allow sub_agent to use external MCP servers (Confluence, Jira, Grafana, etc.), pass their names via the X-MCP-Servers header.

How it works:

Each server URL is constructed from the template: {ANTHROPIC_BASE_URL}/mcp/{server_name} (LiteLLM MCP proxy)
Authorization: ANTHROPIC_AUTH_TOKEN from the container env — no need to pass keys separately
User email from X-User-Email is forwarded to each MCP server

Available servers: check your LiteLLM instance for the list of registered MCP servers.

Format: comma-separated server names (as registered in LiteLLM):

X-MCP-Servers: confluence,jira

Example: sub_agent with Confluence and Jira MCP servers:

curl -s -X POST "http://localhost:8081/mcp/docker_ai" \
  -H "Authorization: Bearer <LITELLM_API_KEY>" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: <SESSION_ID>" \
  -H "X-Chat-Id: my-session" \
  -H "X-MCP-Servers: confluence,jira" \
  -H "X-User-Email: user@example.com" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"sub_agent","arguments":{"task":"Find the onboarding page in Confluence","description":"search confluence"}}}'

Result: before launching Claude Code inside the container, a ~/.mcp.json file is created:

{
  "mcpServers": {
    "confluence": {
      "type": "http",
      "url": "http://localhost:8081/mcp/confluence",
      "headers": {
        "Authorization": "Bearer <ANTHROPIC_AUTH_TOKEN from env>",
        "x-openwebui-user-email": "user@example.com"
      }
    },
    "jira": { ... }
  }
}

If the X-MCP-Servers header is not provided, sub_agent works without MCP servers (as before).

System Prompt for Integrations

A ready-made system prompt that teaches the AI how to work with the Computer Use VM (files, skills, tools). Use it in n8n, your own agents, or any integrations.

Get the Prompt

# Template with placeholders {file_base_url} and {archive_url}
curl http://localhost:8081/system-prompt

# With substituted URLs (for a specific chat)
curl "http://localhost:8081/system-prompt?file_base_url=http://localhost:8081/files/MY_CHAT_ID&archive_url=http://localhost:8081/files/MY_CHAT_ID/archive"

Usage in n8n

Add an HTTP Request node with GET /system-prompt and insert the result into the system message of your AI agent. Substitute file_base_url and archive_url via query params or replace the placeholders in n8n.

File API

GET /files/{chat_id}/{filename}

Download a file from the container's outputs directory

GET /files/{chat_id}/archive

Download all files as a ZIP archive

POST /api/uploads/{chat_id}/{filename}

Upload a file to the container's uploads directory

GET /api/uploads/{chat_id}/manifest

Get the manifest of uploaded files (filename to MD5 mapping)

Direct Access to /mcp (for developers)

Warning: Direct access to localhost:8081/mcp requires a special API key. For normal usage, use the LiteLLM endpoint above.

Example (internal use only)

curl -X POST http://localhost:8081/mcp \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <INTERNAL_MCP_API_KEY>" \
  -H "X-Chat-Id: my-session-123" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"bash_tool","arguments":{"command":"echo Hello","description":"test"}}}'

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        Docker Host                              │
│                                                                 │
│  ┌──────────────────┐      ┌─────────────────────────────────┐ │
│  │  File Server     │      │  Docker Containers              │ │
│  │  (this service)  │      │                                 │ │
│  │                  │      │  ┌─────────────────────────┐   │ │
│  │  POST /mcp ──────┼──────┼──│ owui-chat-{chat_id}     │   │ │
│  │                  │      │  │  - bash_tool            │   │ │
│  │  GET /files/* ───┼──────┼──│  - view                 │   │ │
│  │                  │      │  │  - create_file          │   │ │
│  │                  │      │  │  - str_replace          │   │ │
│  └──────────────────┘      │  └─────────────────────────┘   │ │
│           │                │                                 │ │
│           │                │  ┌─────────────────────────┐   │ │
│           ▼                │  │ owui-chat-{other_id}    │   │ │
│  /var/run/docker.sock      │  └─────────────────────────┘   │ │
│                            └─────────────────────────────────┘ │
│                                                                 │
│  /data/{chat_id}/                                              │
│    ├── uploads/   (files from user)                            │
│    └── outputs/   (files created by AI)                        │
└─────────────────────────────────────────────────────────────────┘