T-FLEX CAD 17 MCP
Read-only MCP server providing search and retrieval of T-FLEX CAD 17 API symbols, documentation, and capabilities via a static API layer.
README
T-FLEX CAD 17 MCP
Read-only Model Context Protocol server over the canonical T-FLEX-CAD-17-API knowledge/API layer.
GitHub Pages publishes only the static project dashboard. The executable MCP runtime is the Node.js/Docker service from this repository and must run on a compute host.
Architecture
flowchart LR
C[MCP client / AI agent]
M[T-FLEX-CAD-17-MCP\nstdio or Streamable HTTP]
A[T-FLEX-CAD-17-API\nraw GitHub content]
P[T-FLEX-CAD-17-MCP Pages\ndocs and diagnostics]
C -->|MCP tools| M
M -->|read-only HTTPS| A
P -->|manifest diagnostics| A
P -.->|health/readiness check| M
The runtime never modifies T-FLEX-CAD-17-API. It reads only the configured upstream base URL and allowlisted documentation paths.
MCP tools
| Tool | Purpose |
|---|---|
tflex_manifest |
Return the dataset and graph manifests. |
tflex_search_symbols |
Search API symbols with optional assembly and kind filters. |
tflex_get_symbol |
Resolve an exact symbol identifier/name, then fall back to the best match. |
tflex_search_docs |
Search normalized CHM/help topic records. |
tflex_search_capabilities |
Search the curated capability-to-symbol/topic seed. |
tflex_get_document |
Read an allowlisted Markdown, CHM mirror, XML API, graph or manifest file. |
All tools are declared read-only and idempotent.
Runtime endpoints
When npm start or the Docker image is running:
- MCP:
POST/GET/DELETE /mcp - Liveness:
GET /healthz - Upstream readiness:
GET /readyz - Service metadata:
GET /
The HTTP transport maintains MCP sessions in process memory and returns MCP-Session-Id after initialization.
Local stdio
npm install
npm run start:stdio
Example client configuration:
{
"mcpServers": {
"t-flex-cad-17": {
"command": "node",
"args": ["/absolute/path/T-FLEX-CAD-17-MCP/src/stdio.js"]
}
}
}
Local Streamable HTTP
npm install
npm start
Default endpoint:
http://localhost:3000/mcp
The runtime also accepts platform-provided PORT; MCP_PORT takes precedence.
Docker
docker build -t t-flex-cad-17-mcp .
docker run --rm -p 3000:3000 t-flex-cad-17-mcp
Or:
docker compose up -d --build
Deploy to a public runtime
Deployment descriptors are included in the repository:
render.yaml— Render Blueprint using the Docker runtime and/healthzhealth check.railway.toml— Railway Docker deployment with/healthzhealth check and restart policy.compose.yaml— VPS/on-premises Docker Compose deployment..github/workflows/container.yml— publishesghcr.io/krickmanc/t-flex-cad-17-mcp:latestand an immutable SHA tag.
After deployment, point the MCP client to:
https://<runtime-host>/mcp
The GitHub Pages address is not the MCP endpoint.
Configuration
| Variable | Default | Meaning |
|---|---|---|
TFLEX_API_BASE_URL |
raw T-FLEX-CAD-17-API/main |
Canonical read-only upstream. |
TFLEX_FETCH_TIMEOUT_MS |
45000 |
Upstream request timeout. |
TFLEX_CACHE_TTL_MS |
300000 |
In-memory dataset cache lifetime. |
TFLEX_MAX_DATASET_BYTES |
67108864 |
Maximum index response size. |
TFLEX_MAX_DOCUMENT_BYTES |
2097152 |
Maximum single document size. |
MCP_MAX_TOOL_OUTPUT_CHARS |
120000 |
Maximum serialized tool result. |
MCP_HOST |
0.0.0.0 |
HTTP bind address. |
MCP_PORT |
3000 |
HTTP port. |
PORT |
unset | Hosting-platform port fallback. |
MCP_ALLOWED_ORIGINS |
* |
Comma-separated CORS origins. |
Verification
npm run check
npm test
npm run smoke:live
CI checks JavaScript syntax, stdio MCP, Streamable HTTP MCP, the live knowledge/API layer, and the Docker image.
GitHub Pages
Deploy GitHub Pages publishes the site/ directory. The page reads the live manifests directly from T-FLEX-CAD-17-API and can probe /healthz and /readyz of a separately deployed runtime.
Current canonical dataset
The current T-FLEX-CAD-17-API manifests report:
- 4 API assemblies
- 2,452 generated type pages
- 17,929 symbols
- 19,350 CHM/help pages
- 38,051 graph nodes
- 51,105 graph edges
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。