For AI Agents
This page is for all AI coding tools (Claude Code, Cursor, Codex, VS Code, Qoder, Trae, Windsurf, Gemini CLI, and any custom harness). It explains how to give your Agent full access to the TTLock Open Platform Cloud API documentation: live API lookup, an offline knowledge pack, and structured full-text data.
1. Copy This Prompt (fastest way)
No configuration needed. Paste this into your Agent conversation or automation workflow:
You will integrate with the TTLock Open Platform Cloud API. Note:
1. API parameters and API domains may differ from your training data — always verify before writing code.
2. First read https://cnopen.ttlock.com/llms.txt for the documentation index and API domain.
3. If you support MCP, connect to https://cnopen.ttlock.com/mcp (Streamable HTTP, no install, no auth)
to look up API parameters and error codes and to generate curl commands in real time.
4. If you can install skills, run: npx skills add -g https://cnopen.ttlock.com/downloads/opendoc-skill-latest.tgz
5. Red line: unlock / power / water control APIs operate real devices — never call them for real; output curl examples only.2. What the Agent Gets
MCP Server — Live Documentation (recommended)
MCP (Model Context Protocol) is an open protocol connecting AI models to external tools and data. We provide a remote HTTP service: no install, no auth, no credentials, synced with the doc site:
https://cnopen.ttlock.com/mcpIt ships a documentation query toolset — API search, API detail, OpenAPI catalog, error-code lookup, curl example generation — invoked automatically by the Agent while you talk in natural language.
Typical conversations after setup:
| You ask | The Agent does |
|---|---|
| “Which APIs are related to eKeys?” | Searches docs, returns the API list with doc links |
| “Give me a curl for listing locks” | Generates a ready-to-run curl (domain and required params included) |
| “What does error 10004 mean?” | Looks up the error table, gives cause and fix |
| “Full parameter table of the remote unlock API” | Returns parameter/response definitions |
| “What API categories do you have?” | Returns the category catalog for follow-up questions |
Skill — Offline Knowledge Pack (intranet / zero latency)
An offline Markdown snapshot of the full API docs: works offline, no need to repeat platform context, real parameter tables reduce API hallucinations:
npx skills add -g https://cnopen.ttlock.com/downloads/opendoc-skill-latest.tgzOnce installed, the Agent follows a three-step workflow: read references/index.md to find the API → read its doc for the parameter table and curl → build requests with the API domain baked into the pack.
Skills load on demand and don't consume resident context: at startup only ~100 tokens of metadata enter the context (the Agent merely knows the skill exists); instructions load when triggered; individual API docs are read only when used.
The Agent uses it automatically when you write integration code; for stable results you can also trigger it explicitly:
- “use the opendoc skill, write code to list my locks”
- “use the opendoc skill, show the parameter table of the send-eKey API”
- “use the opendoc skill, how do I handle error 10004?”
Manage: npx skills list to view installed skills; update by re-running the install command above (it overwrites in place); npx skills remove to uninstall.
Manual install without Node (Claude Code-like clients):
mkdir -p ~/.claude/skills && curl -sL https://cnopen.ttlock.com/downloads/opendoc-skill-latest.tgz | tar -xz -C ~/.claude/skills/ --strip-components=2 package/skills/opendocLLMs.txt — Structured Docs to Feed the Model
Install nothing; feed the address to your model or knowledge base:
| File | Description |
|---|---|
/llms.txt | Navigation file (llmstxt.org spec): site intro, API domain, links to all pages |
/document-pages/llms-full.txt | Full doc text (Chinese first, then English) for one-shot long-context feeding |
How to use in AI tools (shown with llms.txt; swap in llms-full.txt the same way):
| Tool | Usage |
|---|---|
| Cursor | Add the file to project docs via the @Docs feature |
| Windsurf | Reference with @ in chat, or write it into .windsurf/rules |
| Claude Code | Add the URL to workspace docs (Docs / Context Files) |
| Gemini CLI | --context flag or .gemini/config.json |
| Trae | Put it into project knowledge sources and enable referencing |
| Qoder | Add as an external knowledge file in .qoder/config.yml, or reference with @docs in chat |
| Others | Any LLMs.txt-aware tool — just point it at the addresses above |
3. Client Configuration
MCP has a single integration form: a remote HTTP URL. Copy the addresses below as-is; install the Skill with the npx skills add command above.
Claude Code
Option 1, CLI (current project by default; add -s user for user-level, all projects):
claude mcp add --transport http ttlockOpendocMcp https://cnopen.ttlock.com/mcpOption 2, create .mcp.json in the project root (commit it for the whole team):
{
"mcpServers": {
"ttlockOpendocMcp": {
"type": "http",
"url": "https://cnopen.ttlock.com/mcp"
}
}
}Verify: restart the session, run /mcp, and ttlockOpendocMcp should show as connected. Skill location: -g installs to ~/.claude/skills/opendoc/, without -g to the project's .claude/skills/; natively recognized by Claude Code.
Cursor
Project-level: create .cursor/mcp.json in the project root; global (all projects): ~/.cursor/mcp.json. Same content:
{
"mcpServers": {
"ttlockOpendocMcp": {
"url": "https://cnopen.ttlock.com/mcp"
}
}
}Cursor connects automatically after saving. UI route: Settings → MCP → Add Server, type Streamable HTTP / URL, paste the address, and make sure the toggle for ttlockOpendocMcp is enabled (green).
Codex (OpenAI Codex CLI)
Edit ~/.codex/config.toml (create it if missing) and append:
[mcp_servers.ttlockOpendocMcp]
url = "https://cnopen.ttlock.com/mcp"Verify: run codex mcp list; ttlockOpendocMcp should be listed. Skills: if your Codex version has no Skills directory, drop the pack's references/ into the workspace and add one line to AGENTS.md: "When integrating with the TTLock Open Platform Cloud API, first read references/index.md to find the API, then read its doc."
Qoder
UI route: Settings → MCP → Add Server, type Streamable HTTP, enter https://cnopen.ttlock.com/mcp, save and enable. If your version supports JSON config/import, use the common structure:
{
"mcpServers": {
"ttlockOpendocMcp": { "url": "https://cnopen.ttlock.com/mcp" }
}
}Skills: put references/ into the project and add a guiding rule under .qoder/rules/ (same wording as the Codex AGENTS.md example).
Trae
UI route: open the MCP panel → Add → type Streamable HTTP, enter https://cnopen.ttlock.com/mcp, enable it in the list. For JSON import, the structure is the same as Qoder's (see above). Skills: put references/ into the project and add a guiding rule in .trae/rules/project_rules.md.
VS Code
Create .vscode/mcp.json in the project root (note the key is servers, not mcpServers like other clients):
{
"servers": {
"ttlockOpendocMcp": {
"type": "http",
"url": "https://cnopen.ttlock.com/mcp"
}
}
}UI route: command palette → MCP: Add Server, type HTTP, enter the address. Verify with MCP: List Servers.
Windsurf
~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"ttlockOpendocMcp": { "url": "https://cnopen.ttlock.com/mcp" }
}
}Gemini CLI
Configure MCP with the same address (Streamable HTTP): https://cnopen.ttlock.com/mcp.
DeepSeek and custom harnesses
Any harness supporting MCP Streamable HTTP (regardless of the underlying model) integrates the same way — treat the address as a standard MCP Server:
- Endpoint:
https://cnopen.ttlock.com/mcp - Protocol: Streamable HTTP, stateless, no sticky sessions, no auth headers
Common mcpServers JSON (most harnesses accept this structure as-is):
{
"mcpServers": {
"ttlockOpendocMcp": { "url": "https://cnopen.ttlock.com/mcp" }
}
}If the harness doesn't support MCP, feed https://cnopen.ttlock.com/document-pages/llms-full.txt directly, or mount the Skill's references/ directory as a knowledge base.
4. FAQ
Q: The docs were just updated but the Agent answers with old content? MCP data follows within its 24h cache; restart the session if urgent; for the Skill, re-run the install command to update.
Q: Should I put client_id / password into the MCP or Skill? No — and please don't. Both are read-only public documentation services with no secrets; real API calls are made from your own backend as documented.
Q: Could the Agent actually unlock my lock? No. The remote service only provides documentation query tools and nothing that operates real devices.
Q: A Skill/MCP installed from another brand's site doesn't match? Brand wording and API domains are distributed per site. Reinstall once with this site's address to align.
Q: Which option should I install? MCP for daily development (live data); Skill for intranet/offline; feed llms-full.txt for one-off tasks. They can be combined — the Agent picks per scenario.
