seline

Seline ships with built-in MCP support for per-agent servers. This means you can connect external AI services and tooling to individual agents via MCP, all within the desktop application. The MCP servers are bundled with the app for a smooth setup, so you don’t need a separate Node.js installation on supported platforms. When you start Seline, it will auto-connect configured MCP servers per agent, enabling dynamic routing of prompts, tools, and context between your agents and external models or services. You can leverage the multi-model routing to assign different providers or pathways to specific tasks, all while keeping data on-device when desired.

To use the MCP capability, ensure the MCP server is enabled for the agent you’re working with. The server name in your MCP configuration corresponds to the agent or workflow you’re extending. You can customize dynamic variables and per-agent paths to point to your preferred models or tools. For example, you can route a research task to a Claude/DeepRouter combo, or hook up a local vector store for fast retrieval via LanceDB. The integration supports dynamic variables like ${SYNCED_FOLDER} and ${SYNCED_FOLDERS} to reference the agent’s workspace, making it easy to attach files, index documents, or perform local web scraping via the embedded Puppeteer browser for context-enhanced results.

If you need to manage channels or agents, go to the MCP section in the app’s UI to configure per-agent servers, adjust environment variables, and monitor live progress of tool calls and responses. The built-in tooling also includes command execution within synced folders, enabling secure automation workflows without leaving the app.

Prerequisites

• Operating system: Windows, macOS, or Linux
• Node.js 20+ (recommended for Electron environments; 22+ may be needed for native module rebuilds on Electron 39)
• npm 9+ (included with Node.js)

Installation steps

1. Ensure Node.js and npm are installed. Verify versions: node -v npm -v

2. Install the MCP server package (using npm or npx, depending on your setup). Recommended: use the bundled MCP server via npx as shown in the mcp_config. If you prefer local installation:

   npm install -g terCumantanumut-seline-mcp

3. Install local dependencies (if you have a local checkout of the Seline MCP server):

   npm install

4. Start the app or MCP server via the MCP configuration. If running through the MCP config example, you can trigger it via the app or using the npx command defined in mcp_config:

   npx -y terCumantanumut-seline

5. Optional: set up environment secrets and configuration in a .env file as documented by the project (see Required Runtime Secrets in the README).

6. Launch Seline and verify the MCP servers initialize per-agent. Monitor the UI for connection status and log any errors for troubleshooting.

• The MCP servers are designed to work with per-agent configurations. Each agent can have its own server configuration and dynamic variables such as ${SYNCED_FOLDER} or ${SYNCED_FOLDERS}.
• If you encounter sign-in or signing issues on macOS (DMG builds), check for newer workaround discussions or sign in with an updated Apple developer ID as noted in the Known Issues.
• When using local vector search with LanceDB, ensure your embeddings provider (local or API) is configured per agent.
• For tool discovery and multi-model routing, use the 40+ tools loaded on-demand via searchTools to reduce token usage and latency.
• If you rely on Remotion or other media tooling, configure required tokens (REMOTION_MEDIA_TOKEN) in the .env file as part of runtime secrets.
• The MCP feature supports dynamic path variables for per-agent connections; review the Dynamic Variables section in the README to tailor your environment.