askimo

Askimo provides experimental support for Model Context Protocol (MCP) integrations, allowing you to connect and orchestrate external tools and services directly from the Askimo environment. This enables you to define custom tools that can be invoked by AI models during a session, pass context from your conversations to those tools, and retrieve structured results to enrich responses. The MCP integration is designed to be lightweight and local, keeping your data on-device while enabling powerful automations and workflows.

To use MCP with Askimo, first ensure the MCP feature is enabled in the application settings. Then configure your external tools by supplying a manifest that describes each tool's capabilities, input parameters, and authentication requirements. Askimo will load these tool definitions and expose them to the AI models within a single unified interface. When a model needs to perform a task that matches a tool’s capability (for example, running a code analysis, querying a local database, or invoking a file-processing service), Askimo can call the appropriate tool, pass the necessary context, and incorporate the tool’s output back into the chat or workflow.

The included tooling supports orchestration across multiple MCP tools in a single prompt. This means you can chain actions (tool A analyzes a file, tool B fetches related metadata, tool C formats the result) to build sophisticated workflows without leaving the Askimo environment. Keep in mind that MCP is experimental, so you may want to start with a small set of trusted tools and iterate as you validate results and performance.

Prerequisites:

• Ensure you're running Askimo (Desktop) v1.x or later, available for macOS, Windows, and Linux.
• Local development prerequisites are not required to use MCP with the official Askimo desktop, but if you plan to run a separate MCP server implementation locally, you may need Python 3.9+ and related dependencies.

1. Install Askimo Desktop

• Download the latest Askimo desktop installer from the official site: https://askimo.chat/download/
• Run the installer for your OS and complete the setup.

2. Enable MCP experimental support

• Open Askimo > Settings > Experimental Features (or Developer Options).
• Enable MCP integrations and note the default local port or socket used for tool communication.

3. Prepare your MCP tool manifest (local or remote)

• Create a tool manifest (JSON or YAML) describing each tool, its inputs, outputs, and any authentication requirements.
• Examples and schema are available in the Askimo documentation under MCP tools configuration.

4. Start or configure your MCP server (if using a custom server)

• If you’re running a separate MCP server, start it according to its own documentation. Typical setups may include:
  • Python-based MCP server: python -m your_mcp_server
  • Node-based MCP server: npm run mcp-server
• Ensure the server is reachable by Askimo (same machine or accessible network address).

5. Connect Askimo to the MCP server

• In Askimo, provide the server endpoint (host:port) and any required API keys or tokens in the MCP configuration panel.
• Load the tool manifest so Askimo can discover available tools.

Prerequisites recap:

• Askimo desktop app installed
• MCP feature enabled in settings
• Access to an MCP server or manifest definitions for local tooling
• Network access between Askimo and the MCP server if using a separate process

Tips and notes for working with MCP in Askimo:

• Experimental feature: expect occasional API changes in MCP tooling and manifests; keep manifests versioned.
• Local-first privacy: MCP tools can run locally on your machine; ensure manifests and data stay within your chosen scope.
• Environment variables: use ASKIMO_MCP_CONFIG_PATH to point to your tool configuration file and ASKIMO_MCP_LOG_LEVEL to control verbosity for troubleshooting.
• Tool manifests should clearly define input types, required authentication, and expected outputs to prevent ambiguous prompts.
• If you encounter connectivity issues, verify firewall rules and that the MCP server is listening on the correct port.
• Start with a small set of trusted tools and gradually add more capabilities as you validate results and performance.