mcpserver

This MCP server aggregates four powerful platforms into a single multimodal AI toolbox: Bigmodel (GLM), Pollinations.AI, SiYuan (SiYuan Notes), and a web search tool. When you start the server, it exposes MCP endpoints that other clients can invoke to perform image/video analysis, generation and reasoning tasks, read and update content in SiYuan notes, and perform web searches with caching and safety filters. You can pass tool-specific prompts and parameters (for example, image paths, prompts, model names, or search queries) and receive structured results with references. The system is designed to work in a modular way, so you can leverage GLM image/video understanding, CogView image generation, multiple Pollinations.AI models for image/text/audio generation, and advanced web search capabilities with Sonar-based reasoning and citations.

Prerequisites: Node.js (>= 18) and pnpm. Optional: an MCP environment if you’re integrating with other MCP tooling.

1. Clone the repository and navigate into the project folder.
2. Install dependencies:

pnpm install

3. Configure API keys and optional parameters. Create an environment file or export variables, for example:

export GLM_API_KEY=your_glm_api_key_here
export PERPLEXITY_API_KEY=your_perplexity_api_key_here
export SIYUAN_API_TOKEN=your_siyuan_token_here
export SIYUAN_API_BASE=http://127.0.0.1:6806

4. Build the project (if a build step exists):

pnpm build

5. Start the MCP server:

pnpm start

6. Optional: run in MCP mode directly from the built entrypoint:

./build/index.js

Tips and caveats:

• Ensure your GLM, Perplexity, and SiYuan tokens are valid and have the required permissions for the actions you intend to perform.
• The environment variable priorities are: system env > .env in the working directory > project-level .env, so placing keys in a local .env can help with local development.
• For web search, consider adjusting WEB_SEARCH_CACHE_TTL and rate limit settings to balance latency and cost.
• If you encounter module import issues in development mode, use pnpm dev which runs with ts-node/esm as noted in the documentation.
• Check logs regularly; the server exposes logs for both normal operation and errors, which can help diagnose missing keys or misconfigurations.