Enables creation and management of cloud agents that work on GitHub repositories, including launching tasks, auto-creating pull requests, monitoring task status, and managing conversations with agents working on codebases.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Cloud Agent MCP Servercreate a task to fix the login bug in our GitHub repo"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Cloud Agent MCP Server
MCP server for the Cursor Cloud Agents API. Lets AI assistants create and manage cloud agents that work on GitHub repositories.
Quick Start
Installation
Install from npm
Or install locally in your project:
Install from Source
If you're developing or want to run from source:
Configuration
Environment Variables
Variable | Required | Description |
| Yes | API key from cursor.com/settings |
| No | Server port for HTTP version only (default: 3000) |
Connecting Clients
Cursor
Option 1: Using npm package (Recommended)
After installing via npm, create .cursor/mcp.json:
Or if installed globally:
Option 2: From source (Development)
If running from source, create .cursor/mcp.json:
Option 3: HTTP Server (Alternative)
If you prefer the HTTP version, configure it as:
Then run npm start in a separate terminal to start the HTTP server.
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
MCP Inspector
Usage
Typical Workflow
Available Tools
Discovery Tools
Tool | Description |
| Detects current git repo (repository URL, branch, uncommitted changes). Optionally lists all accessible repos. Call this first. Note: When using |
| Get API key info (verify authentication) |
| List available LLM models |
Task Lifecycle Tools
Tool | Description |
| Launch a cloud task with a task prompt |
| List all your cloud tasks |
| Get status of a specific task |
| Send additional instructions to a running task |
| Get full conversation history |
| Permanently delete a task |
Common Examples
Create a Task
List Tasks
Get Repository Info
Reference
Tool Documentation
get_repos - Get Repositories
Basic usage (current repo only):
Fetch all accessible repositories (filters REQUIRED):
Important: When using include_all: true, you MUST provide regex_patterns to filter the results. This prevents returning too many repositories.
Filter repositories with regex patterns:
Multiple filter patterns (OR logic):
Specify working directory:
Filter Examples:
["^my-.*"]- Repos starting with "my-"[".*api.*"]- Repos containing "api"["github.com/myorg"]- Repos from specific org[".*backend.*", ".*server.*"]- Repos matching either pattern
get_me - Verify Authentication
get_models - List Available Models
create_task - Launch a Cloud Task
Basic usage:
With branch specification:
Auto-create PR:
Custom branch name:
Specify model:
With plan file:
Complete example with all options:
list_tasks - List All Tasks
Basic listing:
Limit results:
Pagination with cursor:
Filter by status:
Filter by repository:
Filter by branch name:
Filter by summary content:
Combine filters:
Filter Examples:
"FINISHED"- Only finished tasks"RUNNING|CREATING"- Active tasks".*api.*"- Tasks working on API repos"feature/.*"- Tasks on feature branches"FINISHED.*README"- Finished tasks with README in summary"FAILED|CANCELLED"- Failed or cancelled tasks
get_task - Get Task Status
add_followup - Send Follow-up Instructions
get_conversation - Get Conversation History
delete_task - Delete a Task
Response Shapes
All tools return structured JSON responses matching the Cloud Agents API specification. See docs.md for complete API documentation.
get_me Response
get_models Response
get_repos Response
create_task Response
list_tasks Response
get_task Response
Status Values: CREATING, RUNNING, FINISHED, FAILED, CANCELLED
get_conversation Response
add_followup Response
delete_task Response
Advanced
Plan Parallel Tasks
Use /plan-parallel-tasks to break down a project for multiple parallel tasks:
This will:
Call
get_reposto detect your repositoryBreak the project into independent tasks
Group tasks into phases (parallel → sequential)
Provide exact prompts for each
create_taskcall
Parallelization Rules
CAN run in parallel:
Tasks that modify completely different files
Tasks that create new files without touching shared files
CANNOT run in parallel:
Tasks that modify the same file
Tasks where one depends on another's output
Tasks that both modify package.json, tsconfig.json, etc.
Server Versions
This package includes two server versions:
HTTP Server (
src/server.ts): Runs an Express HTTP server on port 3000. Use for remote connections or when you need HTTP endpoints.Stdio Server (
src/server-stdio.ts): Uses standard input/output. Recommended for local integrations. Better for process-based spawning.
Run with:
HTTP:
npm start(default)Stdio:
npm run start:stdio
Development
Health Check
The server exposes a health check endpoint:
Returns: {"status":"ok","service":"cursor-cloud-agent-mcp","version":"1.0.2"}
Publishing to npm
To publish this package to npm:
The package will be available as cursor-cloud-agent-mcp on npm.
License
MIT