Cangje Docs

You set up a local MCP server to power Claude Code’s document search for the Cangjie language. The server exposes a structured, MCP-based interface that lets Claude Code efficiently query Cangjie documentation, speeding up learning and reference tasks.

How to use

Interact with the server through Claude Code by asking natural language questions. You can perform foundational lookups like “please help me find the syntax for variable declarations in Cangjie” or topic-driven searches like “I want to understand the basic data types in Cangjie.” The server will reply with relevant documentation sections, examples, and API references directly drawn from your local corpus.

How to install

Prerequisites depend on the MCP client you use, but you will typically need a working environment capable of running an MCP server and a client that can connect to it. The following steps outline a concrete flow based on the provided setup.

Clone the document corpus and prepare the MCP runtime as shown in the configuration example.

Step 1. Clone the document corpus repository (or place your existing corpus in the expected directory):

git clone https://gitcode.com/Cangjie/CangjieCorpus.git

Step 2. Ensure the document directory exists in the path you will reference in the MCP configuration, or use the -dir option to specify it:

# Ensure a CangjieCorpus directory is present
# Or specify the directory when starting the MCP

Step 3. Add the MCP configuration for Claude Code using the provided snippet:

{ "mcpServers": { "cangjie-docs": { "command": "/path/to/cangje-docs-mcp", "args": ["-dir", "/path/to/CangjieCorpus"] } } }

Step 4. Restart Claude Code to load the new MCP configuration.

## Additional sections

Configuration details you need are shown in the MCP setup blocks. For reference, a compliant local MCP configuration should look like this when you verify it in your environment:

{ "mcpServers": { "cangje-docs": { "type": "stdio", "command": "/path/to/cangje-docs-mcp", "args": ["-dir", "/path/to/CangjieCorpus"], "env": {} } } }

## Troubleshooting

Common issues you might run into and how to address them:

Q: Claude Code can’t find the MCP server
A: Verify the path in the configuration is correct and the executable has permission to run.

Q: Search results are empty
A: Confirm the document directory exists and contains Cangjie documents, and check directory permissions.

Q: Server starts but reports the document directory does not exist
A: Use the -dir parameter to point to the correct directory.

Q: MCP server fails to start
A: Ensure the configuration format is correct and the provided command and arguments match what the server expects.

{ "mcpServers": { "cangje-docs": { "type": "stdio", "command": "/path/to/cangje-docs-mcp", "args": ["-dir", "/path/to/CangjieCorpus"], "env": {} } } }

## 文档目录结构

Your local corpus should follow a structured layout to keep content organized and easily searchable. A typical layout looks like this:

CangjieCorpus/ ├── README.md ├── manual/ # Basic manuals │ ├── first_understanding/ │ ├── basic_data_type/ │ ├── function/ │ └── ... ├── libs/ # Standard library APIs │ └── std/ ├── tools/ # Development tools └── extra/ # Additional content

## 开始使用

配置完成后，你可以在 Claude Code 中自然地查询仓颉语言的所有文档内容。系统会自动理解你的问题并提供相关的文档内容和建议。