MCP Service Address
mcp: https://cyeam-wiki-mcp-production.up.railway.app/mcp
Standard Process (Correct Way to Use MCP)
1. Discovery: Connect to the MCP service to get the capability list
After connecting, discovery happens automatically, yielding the following three categories of capabilities:
- Prompts list
- Tools list
- Resources list
2. Call the MCP Prompt:wiki_query_system
This is the corevector group prompt, call it directly as the system prompt, containing:
- Architect role setting
- Knowledge base retrieval rules
- Output format requirements
3. Call MCP Wiki Tools to retrieve context
Based on the user's question, call the following tools in priority order to obtain relevant content:
| Tool name | Parameters |
|---|---|
wiki_query |
{"question": q, "depth": 2} |
wiki_search_index |
{"keyword": q} |
wiki_get_graph |
{"query": q, "depth": 1, "max_nodes": 50} |
Follow-up action: Extract from the tool's return result the [[文章名]] marker, call wiki_get_article to get the full article content.
4. Real-time tool invocation
When the question needs real-time data:
- Find the
curlorwebFetchtool - Generate the URL to visit (without the https:// prefix)
- Call the tool:
{"url": "生成的URL"}
5. Integrate all MCP context to answer
- System prompt (from
wiki_query_systemprompt) - Wiki tool retrieval results
- (Optional) real-time tool return data
- Reference links must be included at the end
MCP Capability Reference (Discovery results, use as needed)
Prompts (call this first):
wiki_query_system- Architect knowledge base query system prompt
Tools (call as needed):
wiki_query- Deep Q&Awiki_search_index- Keyword searchwiki_get_graph- Knowledge graphwiki_get_article- Get article detailscurl- Real-time HTTP request
Resources (read only when referenced in tool returns, not right away):
wiki://index- The master index of all wiki articleswiki://backlinks- Wiki Backlinks, article backlink index