Skip to main content

Cangjie MCP Server

CI PyPI Python Versions GitHub Release License

仓颉编程语言的 MCP (Model Context Protocol) 服务器,提供文档搜索和代码智能功能。

功能

  • 文档搜索: 基于向量检索的仓颉语言文档搜索
  • 代码智能: 基于 LSP 的跳转定义、查找引用、悬停信息等功能
  • 可选远程文档服务: 支持连接远程文档/索引服务,减少本地资源占用,适合开箱即用或团队共享

安装

你可以根据偏好的包管理器,选择通过 PyPI (Python) 或 npm (Node.js) 进行安装。

通过 PyPI 安装 (Python)

pip install cangjie-mcp

安装并指定 Rust feature

当平台没有预编译 wheel,或你希望按需启用 GPU/NPU 等加速后端时,可强制从 sdist 构建并传入 maturin 构建参数:

pip install --no-binary cangjie-mcp cangjie-mcp \
  --config-settings=build-args="--features local"

可用 feature(传给 cangjie-mcp-cli):

  • local:本地向量化(CPU;默认构建的二进制已启用)
  • legacy:本地向量化改用 ort-tract 后端(适配旧 glibc 构建)
  • local-cuda / local-cudnn:启用 CUDA/CUDNN 后端
  • local-metal:启用 Apple Metal 后端
  • local-mkl / local-accelerate:启用 MKL / Accelerate 后端

通过 npm 安装 (Node.js)

npm install -g cangjie-mcp

npm 默认优先使用预编译二进制;当当前平台没有匹配包,或 Linux 的 glibc 低于 2.28 时,会自动回退到本地源码构建。你也可以显式强制源码构建:

CANGJIE_MCP_FORCE_BUILD=1 npm install -g cangjie-mcp

架构

cangjie-mcp 支持两种运行模式:

本地模式(默认)

MCP 服务器在本地加载检索索引(BM25 + 向量索引),直接处理查询。

cangjie-mcp               # 或 uvx cangjie-mcp

远程服务模式(可选)

如果你不想在本机下载/加载向量模型与索引(或希望多人/多台机器共享同一套索引),可以把检索能力独立部署为 cangjie-mcp-server,同时对外提供:

  • HTTP 查询 API,供本地 cangjie-mcp --server-url 连接(LSP 工具仍在本地可用)
  • Remote MCP 端点,MCP 客户端无需本地运行 cangjie-mcp 即可直接连接

注意cangjie-mcp-server 为独立二进制,不随 PyPI / npm 发行,需单独从源码构建。Remote MCP 模式下 LSP 工具(cangjie_lsp)不可用 —— 它依赖本地文件系统和仓颉 SDK,仅在 stdio 模式(cangjie-mcp)中提供。

# 启动服务器(同时暴露 HTTP API、Streamable HTTP MCP、SSE MCP)
cangjie-mcp-server --embedding local --port 8765

# 方式 A:本地 stdio MCP 连接远程索引(保留 LSP)
cangjie-mcp --server-url http://localhost:8765

# 方式 B:客户端直接连接 Remote MCP(无需本地进程)
#   Streamable HTTP: http://localhost:8765/mcp
#   SSE:             http://localhost:8765/sse

快速配置

项目提供两部分能力,可独立安装:MCP 服务器(文档检索 + LSP 工具)和 Skill(仓颉语言速查知识)。推荐搭配使用。

MCP 服务器

公共文档服务器已部署在 https://cj-mcp.learningman.top,直接使用 Streamable HTTP 远程 MCP,无需在本地安装任何包,也无需加载嵌入模型,开箱即用:

npx add-mcp https://cj-mcp.learningman.top/mcp

如需 LSP 代码智能工具(需要本地仓颉 SDK),改用 stdio 模式,通过 uvx 拉起本地进程并连接远程检索服务:

npx add-mcp "uvx cangjie-mcp" \
  --env "CANGJIE_SERVER_URL=https://cj-mcp.learningman.top" \
  --env "CANGJIE_HOME=/path/to/cangjie-sdk"

完全离线 / 自建索引(不连接远程服务器,本地加载嵌入模型):

npx add-mcp "uvx cangjie-mcp" \
  --env "CANGJIE_RERANK_TYPE=local" \
  --env "CANGJIE_HOME=/path/to/cangjie-sdk"

add-mcp 会自动识别当前环境中的 AI 编程助手(Claude Code、Cursor、Windsurf、VS Code、Zed、Claude Desktop 等)并写入对应配置,详见 neondatabase/add-mcp。使用 -g 全局安装,-a <agent> 指定目标,--all 写入所有受支持的客户端。

Skill

本项目提供仓颉语言的 AI 编程助手技能,包含语法速查、关键差异和常见陷阱等知识。项目根目录的 SKILL.md 遵循 skills 规范,可直接通过以下命令安装到你的项目:

npx skills add Zxilly/cangjie-mcp

可用工具

文档搜索

工具名称 功能
cangjie_search_docs 语义搜索仓颉文档

代码智能

需要设置 CANGJIE_HOME 环境变量指向仓颉 SDK 路径,LSP 工具才会注册。

工具名称 功能
cangjie_lsp 统一 LSP 入口,通过 operation 执行 definition、references、hover、document_symbol、diagnostics、workspace_symbol、incoming/outgoing calls 和 type hierarchy

命令行参考

cangjie-mcp

安装 Python 包后,cangjie-mcp 命令可直接使用。

cangjie-mcp                        # 启动 MCP stdio 服务器(无参数时默认行为)
cangjie-mcp query "泛型"           # CLI 搜索(自动启动后台 daemon)
cangjie-mcp lsp hover main.cj --symbol main  # LSP 操作
cangjie-mcp index                  # 构建搜索索引
cangjie-mcp config init            # 生成默认配置文件

cangjie-mcpcangjie-mcp index 接受完整的索引/嵌入/网络选项(通过 cangjie-mcp --help 查看)。其他子命令的设置统一从配置文件加载,运行 cangjie-mcp config path 查看路径。

本地模型缓存

当使用 --embedding localCANGJIE_RERANK_TYPE=local 时,fastembed 模型缓存写入 <data_dir>/cache/fastembed。以下选项同时适用于 cangjie-mcpcangjie-mcp-server

CLI 参数 环境变量 说明
--data-dir PATH CANGJIE_DATA_DIR 应用数据根目录,默认 ~/.cangjie-mcpfastembed 缓存始终为 <data_dir>/cache/fastembed

全局选项

CLI 参数 环境变量 说明
--log-file PATH CANGJIE_LOG_FILE 日志文件路径
--debug CANGJIE_DEBUG 启用调试模式
-h, --help - 显示帮助
-V, --version - 显示版本

环境变量

环境变量 说明
CANGJIE_HOME 仓颉 SDK 路径,设置后自动启用 LSP 工具

cangjie-mcp-server

启动 HTTP 查询服务器和 Remote MCP 服务器,加载本地检索索引(BM25 + 向量索引),通过 HTTP 提供查询服务,同时提供 Streamable HTTP(/mcp)和旧版 SSE(/sse)两种 MCP 传输方式。该服务器为独立二进制,需单独构建。

cangjie-mcp-server [OPTIONS]

支持所有与 cangjie-mcp 相同的索引选项,以及:

CLI 参数 环境变量 默认值 说明
--host TEXT CANGJIE_SERVER_HOST 127.0.0.1 HTTP 服务器监听地址
-p, --port INT CANGJIE_SERVER_PORT 8765 HTTP 服务器监听端口
--mcp-path TEXT CANGJIE_MCP_PATH /mcp Streamable HTTP MCP 端点挂载路径
--no-mcp CANGJIE_NO_MCP - 禁用 Streamable HTTP MCP 端点
--no-sse CANGJIE_NO_SSE - 禁用旧版 SSE 传输端点

HTTP API

方法 路径 说明
GET /health 健康检查
GET /info 索引元数据
POST /search 向量搜索

MCP 端点

默认同时提供两种 MCP 传输方式,暴露 cangjie_search_docs 工具:

传输方式 端点 说明
Streamable HTTP /mcp(可通过 --mcp-path 配置) 现代 MCP 传输,使用 --no-mcp 禁用
SSE(旧版) GET /sse + POST /sse 兼容旧客户端,使用 --no-sse 禁用

注意:Remote MCP 模式下 LSP 工具不可用,仅提供文档搜索相关工具。

Daemon 管理

CLI 工具命令(querylsp 等)会自动在后台启动 daemon 进程,复用已初始化的服务实例以加速响应。daemon 空闲超时后自动退出。

cangjie-mcp daemon status           # 查看 daemon 状态
cangjie-mcp daemon stop             # 停止 daemon
cangjie-mcp daemon logs --tail 50   # 查看日志
cangjie-mcp daemon logs --follow    # 实时跟踪日志

调试与日志

--log-file--debug 配合使用,可以帮助排查 MCP 通信问题:

# 记录应用日志到文件
cangjie-mcp --log-file /tmp/cangjie.log

# 调试模式:额外记录 MCP stdio 协议流量
cangjie-mcp --log-file /tmp/cangjie.log --debug

# 通过环境变量配置
CANGJIE_LOG_FILE=/tmp/cangjie.log CANGJIE_DEBUG=1 cangjie-mcp

许可证

MIT License

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

cangjie_mcp-0.5.0.tar.gz (176.7 kB view details)

Uploaded Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

cangjie_mcp-0.5.0-py3-none-win_amd64.whl (24.1 MB view details)

Uploaded Python 3Windows x86-64

cangjie_mcp-0.5.0-py3-none-manylinux_2_39_x86_64.whl (25.0 MB view details)

Uploaded Python 3manylinux: glibc 2.39+ x86-64

cangjie_mcp-0.5.0-py3-none-manylinux_2_39_aarch64.whl (25.5 MB view details)

Uploaded Python 3manylinux: glibc 2.39+ ARM64

cangjie_mcp-0.5.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (23.1 MB view details)

Uploaded Python 3manylinux: glibc 2.17+ x86-64

cangjie_mcp-0.5.0-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (21.8 MB view details)

Uploaded Python 3manylinux: glibc 2.17+ ARM64

cangjie_mcp-0.5.0-py3-none-macosx_11_0_arm64.whl (22.9 MB view details)

Uploaded Python 3macOS 11.0+ ARM64

File details

Details for the file cangjie_mcp-0.5.0.tar.gz.

File metadata

  • Download URL: cangjie_mcp-0.5.0.tar.gz
  • Upload date:
  • Size: 176.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for cangjie_mcp-0.5.0.tar.gz
Algorithm Hash digest
SHA256 979e5d1de8dd7d9221280d1546f6e8d7619a9f400c7e38dd46a339f281f5f8be
MD5 7c6990ac2406b2c250179a039cab3cca
BLAKE2b-256 44aaded08714f86ec5e2064f550eaa3c905c8b660f5490c3a71780f3af947457

See more details on using hashes here.

File details

Details for the file cangjie_mcp-0.5.0-py3-none-win_amd64.whl.

File metadata

  • Download URL: cangjie_mcp-0.5.0-py3-none-win_amd64.whl
  • Upload date:
  • Size: 24.1 MB
  • Tags: Python 3, Windows x86-64
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for cangjie_mcp-0.5.0-py3-none-win_amd64.whl
Algorithm Hash digest
SHA256 8a8214e1181bf2d85c7ce188d15e16cc02ffdd55ec7106c73e922dc177d34e30
MD5 6f7a15a7ebc93c743cb5a8946c0edb7d
BLAKE2b-256 c233f59009b99f30ee3785700771977bd149750fe194d92c595af7a2ffdfd513

See more details on using hashes here.

File details

Details for the file cangjie_mcp-0.5.0-py3-none-manylinux_2_39_x86_64.whl.

File metadata

  • Download URL: cangjie_mcp-0.5.0-py3-none-manylinux_2_39_x86_64.whl
  • Upload date:
  • Size: 25.0 MB
  • Tags: Python 3, manylinux: glibc 2.39+ x86-64
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for cangjie_mcp-0.5.0-py3-none-manylinux_2_39_x86_64.whl
Algorithm Hash digest
SHA256 21e928731f87a41807a3c71485f6994018712c8c0659a88fabf2adbe6ad55c54
MD5 09b7ac74f9515cbaa7948776090fd03d
BLAKE2b-256 2dac6c0704128dabc2186b0626085259d729a6de06d0edee5ecfbef12776e89b

See more details on using hashes here.

File details

Details for the file cangjie_mcp-0.5.0-py3-none-manylinux_2_39_aarch64.whl.

File metadata

  • Download URL: cangjie_mcp-0.5.0-py3-none-manylinux_2_39_aarch64.whl
  • Upload date:
  • Size: 25.5 MB
  • Tags: Python 3, manylinux: glibc 2.39+ ARM64
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for cangjie_mcp-0.5.0-py3-none-manylinux_2_39_aarch64.whl
Algorithm Hash digest
SHA256 adaab41e876d6305aaa98e667d739cdb453d88dac1da22aa791a2bed82257dfb
MD5 6b72921a1386e59fdeb3fb0e61976004
BLAKE2b-256 26756d79fa8484021cee795b7640ce82e9198f6726da53de9837be66ef1e1827

See more details on using hashes here.

File details

Details for the file cangjie_mcp-0.5.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.

File metadata

  • Download URL: cangjie_mcp-0.5.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
  • Upload date:
  • Size: 23.1 MB
  • Tags: Python 3, manylinux: glibc 2.17+ x86-64
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for cangjie_mcp-0.5.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
Algorithm Hash digest
SHA256 14bd2b40994da2dfb019c1e2bce7641bd924cc771abfb58591f7d19674362fb0
MD5 c5c37aae65739eab642d5dfe6cde72aa
BLAKE2b-256 905105873c10e51f68882d5b54e76b72889bde98977b3cc705f64d0972788951

See more details on using hashes here.

File details

Details for the file cangjie_mcp-0.5.0-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.whl.

File metadata

  • Download URL: cangjie_mcp-0.5.0-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
  • Upload date:
  • Size: 21.8 MB
  • Tags: Python 3, manylinux: glibc 2.17+ ARM64
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for cangjie_mcp-0.5.0-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
Algorithm Hash digest
SHA256 e766fb42afba329e9a8d6db80eb711f46fba710f63010605283dd22d3fe95d56
MD5 4e2d9c49c3fed0cde85e91933ba829e5
BLAKE2b-256 8668cc2b0b39aae364dc60805d499a73973c31e0ad886b3d445d387f61e183ae

See more details on using hashes here.

File details

Details for the file cangjie_mcp-0.5.0-py3-none-macosx_11_0_arm64.whl.

File metadata

  • Download URL: cangjie_mcp-0.5.0-py3-none-macosx_11_0_arm64.whl
  • Upload date:
  • Size: 22.9 MB
  • Tags: Python 3, macOS 11.0+ ARM64
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for cangjie_mcp-0.5.0-py3-none-macosx_11_0_arm64.whl
Algorithm Hash digest
SHA256 b3b6276ecd9cc631f6e43726719e41fa721da1aebc6e8eecaadecc279c2a5455
MD5 58ef2373cb0b510da2ce018b89bb9924
BLAKE2b-256 d707512733a10d284b0f2d6a661e6b673aa9ea4aad18b82c5b192754b5e38eab

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page