接入你的智能体
從發布者安裝。不是 PlugMCP 端點。
遠端 URL 還是本機指令? 兩種安裝方式的差異
ATLAS (Adaptive Task & Logic Automation System) is a project, knowledge, and task management system for LLM Agents.
Built on a 3-node architecture:
+-------------------------------------------+
| PROJECT |
|-------------------------------------------|
| id: string |
| name: string |
| description: string |
| status: string |
| urls?: Array<{title: string, url: string}>|
| completionRequirements: string |
| outputFormat: string |
| taskType: string |
| createdAt: string |
| updatedAt: string |
+----------------+--------------------------+
| |
| |
v v
+----------------------------------+ +----------------------------------+
| TASK | | KNOWLEDGE |
|----------------------------------| |----------------------------------|
| id: string | | id: string |
| projectId: string | | projectId: string |
| title: string | | text: string |
| description: string | | tags?: string[] |
| priority: string | | domain: string |
| status: string | | citations?: string[] |
| assignedTo?: string | | createdAt: string |
| urls?: Array<{title: string, | | |
| url: string}> | | updatedAt: string |
| tags?: string[] | | |
| completionRequirements: string | | |
| outputFormat: string | | |
| taskType: string | | |
| createdAt: string | | |
| updatedAt: string | | |
+----------------------------------+ +----------------------------------+
Implemented as a Model Context Protocol (MCP) server, ATLAS allows LLM agents to interact with a project management database, enabling them to manage projects, tasks, and knowledge items.
Important Version Note: Version 1.5.4 is the last version that uses SQLite as the database. Version 2.0 and onwards has been completely rewritten to use Neo4j, which requires either:
- Self-hosting using Docker (docker-compose included in repository)
- Using Neo4j AuraDB cloud service: https://neo4j.com/product/auradb/
Version 2.5.0 introduces a new 3-node system (Projects, Tasks, Knowledge) that replaces the previous structure.
ATLAS implements the Model Context Protocol (MCP), enabling standardized communication between LLMs and external systems through:
The Atlas Platform integrates these components into a cohesive system:
Clone the repository:
git clone https://github.com/cyanheads/atlas-mcp-server.git
cd atlas-mcp-server
Install dependencies:
npm install
Configure Neo4j: Ensure you have a Neo4j instance running and accessible. You can start one using the provided Docker configuration:
docker-compose up -d
Update your .env file with the Neo4j connection details (see Configuration).
Build the project:
npm run build
Most MCP Clients run the server automatically, but you can also run it manually for testing or development purposes using the following commands.
ATLAS MCP Server supports multiple transport mechanisms for communication:
Standard I/O (stdio): This is the default mode and is typically used for direct integration with local MCP clients (like IDE extensions).
npm run start:stdio
This uses the MCP_TRANSPORT_TYPE=stdio setting.
Streamable HTTP: This mode allows the server to listen for MCP requests over HTTP, suitable for remote clients or web-based integrations.
npm run start:http
This uses the MCP_TRANSPORT_TYPE=http setting. The server will listen on the host and port defined in your .env file (e.g., MCP_HTTP_HOST and MCP_HTTP_PORT, defaulting to 127.0.0.1:3010). Ensure your firewall allows connections if accessing remotely.
A basic Web UI is available for viewing Project, Task, & Knowledge details.
Opening the UI:
npm run webui
Functionality:
Environment variables should be set in the client config in your MCP Client, or in a .env file in the project root for local development.
# Neo4j Configuration
NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=password2
# Application Configuration
MCP_LOG_LEVEL=debug # Minimum logging level. Options: emerg, alert, crit, error, warning, notice, info, debug. Default: "debug".
LOGS_DIR=./logs # Directory for log files. Default: "./logs" in project root.
NODE_ENV=development # 'development' or 'production'. Default: "development".
# MCP Transport Configuration
MCP_TRANSPORT_TYPE=stdio # 'stdio' or 'http'. Default: "stdio".
MCP_HTTP_HOST=127.0.0.1 # Host for HTTP transport. Default: "127.0.0.1".
MCP_HTTP_PORT=3010 # Port for HTTP transport. Default: 3010.
# MCP_ALLOWED_ORIGINS=http://localhost:someport,https://your-client.com # Optional: Comma-separated list of allowed origins for HTTP CORS.
# MCP Security Configuration
# MCP_AUTH_SECRET_KEY=your_very_long_and_secure_secret_key_min_32_chars # Optional: Secret key (min 32 chars) for JWT authentication if HTTP transport is used. CRITICAL for production. *Note: Production environment use has not been tested yet.*
MCP_RATE_LIMIT_WINDOW_MS=60000 # Rate limit window in milliseconds. Default: 60000 (1 minute).
MCP_RATE_LIMIT_MAX_REQUESTS=100 # Max requests per window per IP for HTTP transport. Default: 100.
# Database Backup Configuration
BACKUP_MAX_COUNT=10 # Maximum number of backup sets to keep. Default: 10.
BACKUP_FILE_DIR=./atlas-backups # Directory where backup files will be stored (relative to project root). Default: "./atlas-backups".
Refer to src/config/index.ts for all available environment variables, their descriptions, and default values.
How you configure your MCP client depends on the client itself and the chosen transport type. An mcp.json file in the project root can be used by some clients (like mcp-inspector) to define server configurations; update as needed.
For Stdio Transport (Example Configuration):
{
"mcpServers": {
"atlas-mcp-server-stdio": {
"command": "node",
"args": ["/full/path/to/atlas-mcp-server/dist/index.js"],
"env": {
"NEO4J_URI": "bolt://localhost:7687",
"NEO4J_USER": "neo4j",
"NEO4J_PASSWORD": "password2",
"MCP_LOG_LEVEL": "info",
"NODE_ENV": "development",
"MCP_TRANSPORT_TYPE": "stdio"
}
}
}
}
For Streamable HTTP (Example Configuration):
If your client supports connecting to an MCP server via Streamable HTTP, you provide the server's endpoint (e.g., http://localhost:3010/mcp) in your client configuration.
{
"mcpServers": {
"atlas-mcp-server-http": {
"command": "node",
"args": ["/full/path/to/atlas-mcp-server/dist/index.js"],
"env": {
"NEO4J_URI": "bolt://localhost:7687",
"NEO4J_USER": "neo4j",
"NEO4J_PASSWORD": "password2",
"MCP_LOG_LEVEL": "info",
"NODE_ENV": "development",
"MCP_TRANSPORT_TYPE": "http",
"MCP_HTTP_PORT": "3010",
"MCP_HTTP_HOST": "127.0.0.1"
// "MCP_AUTH_SECRET_KEY": "your-secure-token" // If authentication is enabled on the server
}
}
}
}
Note: Always use absolute paths for args when configuring client commands if the server is not in the client's immediate working directory. The MCP_AUTH_SECRET_KEY in the client's env block is illustrative; actual token handling for client-to-server communication would depend on the client's capabilities and the server's authentication mechanism (e.g., sending a JWT in an Authorization header).
The codebase follows a modular structure:
src/
├── config/ # Configuration management (index.ts)
├── index.ts # Main server entry point
├── mcp/ # MCP server implementation (server.ts)
│ ├── resources/ # MCP resource handlers (index.ts, types.ts, knowledge/, projects/, tasks/)
│ └── tools/ # MCP tool handlers (individual tool directories)
├── services/ # Core application services
│ └── neo4j/ # Neo4j database services (index.ts, driver.ts, backupRestoreService.ts, etc.)
├── types/ # Shared TypeScript type definitions (errors.ts, mcp.ts, tool.ts)
└── utils/ # Utility functions and internal services (e.g., logger, errorHandler, sanitization)
ATLAS provides a comprehensive suite of tools for project, task, and knowledge management, callable via the Model Context Protocol.
ATLAS exposes project, task, and knowledge data through standard MCP resource endpoints.
ATLAS provides functionality to back up and restore the Neo4j database content. The core logic resides in src/services/neo4j/backupRestoreService.ts.
Project, Task, and Knowledge nodes, along with their relationships, into separate JSON files. A full-export.json containing all data is also created.atlas-backup-YYYYMMDDHHMMSS) within the configured backup path (default: ./atlas-backups/). This directory contains projects.json, tasks.json, knowledge.json, relationships.json, and full-export.json.npm run db:backup
This command executes src/services/neo4j/backupRestoreService/scripts/db-backup.ts, which calls the exportDatabase function.full-export.json if available.npm run db:import <path_to_backup_directory>
Replace <path_to_backup_directory> with the actual path to the backup folder (e.g., ./atlas-backups/atlas-backup-20250326120000). This command executes src/services/neo4j/backupRestoreService/scripts/db-import.ts, which calls the importDatabase function.id properties stored within the nodes during export. Ensure your nodes have consistent id properties for relationships to be restored correctly.The examples/ directory contains practical examples demonstrating various features of the ATLAS MCP Server.
examples/backup-example/, this shows the structure and format of the JSON files generated by the npm run db:backup command. See the Examples README for more details.examples/deep-research-example/, this demonstrates the output and structure generated by the atlas_deep_research tool. It includes a markdown file (covington_community_grant_research.md) summarizing the research plan and a JSON file (full-export.json) containing the raw data exported from the database after the research plan was created. See the Examples README for more details.Apache License 2.0
A Model Context Protocol (MCP) server for ATLAS, a Neo4j-powered task management system for LLM Agents - implementing a three-tier architecture (Projects, Tasks, Knowledge) to manage complex workflows
總覽
A Model Context Protocol (MCP) server for ATLAS, a Neo4j-powered task management system for LLM Agents - implementing a three-tier architecture (Projects, Tasks, Knowledge) to manage complex workflows. Now with Deep Research.
工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
Atlas 透過 stdio 在本機執行。
上次檢查時間:2026-10-26。 A Model Context Protocol (MCP) server for ATLAS, a Neo4j-powered task management system for LLM Agents - implementing a three-tier architecture (Projects, Tasks, Knowledge) to manage complex workflows. Now with Deep Research.