From mcp-architect
This skill should be used when the user asks for "MCP examples", "real-world patterns", "code search patterns", "browser proxy patterns", "process management patterns", "show me examples", or wants to see actual implementations from lci, agnt, or other real MCPs.
How this skill is triggered — by the user, by Claude, or both
Slash command
/mcp-architect:mcp-examplesThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Provide real-world MCP patterns from production servers: code search (lci), browser integration (agnt), process management, and knowledge bases.
Provide real-world MCP patterns from production servers: code search (lci), browser integration (agnt), process management, and knowledge bases.
search - Hub tool
{
"input": {"pattern": "string", "filter": "optional"},
"output": {
"results": [
{"id": "r1", "name": "User.authenticate", "preview": "...", "conf": 0.95}
],
"has_more": true,
"total": 127
}
}
get_definition - Spoke tool
{
"input": {"id": "r1"},
"output": {
"symbol_id": "s1",
"name": "User.authenticate",
"signature": "...",
"source": "...",
"location": {"file": "user.ts", "line": 42}
}
}
Token efficiency: ID reference saves ~80% tokens vs. repeating full code
Query: "authenticate"
High match (0.95): Full details (200 tokens)
- Name, signature, docs, preview, location
Medium match (0.70): Summary (50 tokens)
- Name, type, file
Low match (0.40): Minimal (10 tokens)
- Name, ID only
proxy_start - Create
{
"input": {"target_url": "http://localhost:3000"},
"output": {
"proxy_id": "dev",
"listen_addr": "http://localhost:12345",
"status": "running"
}
}
currentpage - Aggregation
{
"input": {"proxy_id": "dev"},
"output": {
"session_id": "page-1",
"url": "http://localhost:3000",
"errors_count": 3, // Not full error objects
"interactions_count": 127, // Not every interaction
"mutations_count": 45, // Not every mutation
"performance": {...}
},
"detail_access": "Use detail=['errors'] for full data"
}
Key pattern: Counts in overview, full data on request
proxy_id (dev)
↓
session_id (page-1)
↓
request_id (req_a1b2)
Each level provides more specificity.
Level 1 - Count
{
"active": 5,
"stopped": 2
}
Level 2 - List
{
"processes": [
{"id": "p1", "name": "dev-server", "status": "running"},
{"id": "p2", "name": "test", "status": "running"}
]
}
Level 3 - Status
{
"id": "p1",
"status": "running",
"uptime": "2h15m",
"memory": "245MB",
"preview": "Server listening :3000"
}
Level 4 - Full
{
/* ...all Level 3... */,
"full_output": "... complete logs ...",
"env": {...},
"metrics": {...}
}
list_topics()
→ ["auth", "deploy", "monitor"]
get_topic_summary("auth")
→ {articles: 12, updated: "2024-01"}
search_articles("OAuth")
→ [{id: "a1", title: "...", preview: "..."}]
get_article("a1")
→ {title, content, related: [...]}
All use IDs to avoid repeating data:
Savings: 70-90% token reduction
All vary detail by context:
All include standard flags:
{
"has_more": boolean,
"total": integer,
"returned": integer,
"complete": boolean
}
All accept unknown params with warnings:
const {known, params, ...extra} = input
if (extra) warnings.push(`Unknown: ${Object.keys(extra)}`)
Before (wasteful):
// Tool 1
{"results": [{"name": "...", "code": "... 200 lines ..."}]}
// Tool 2 needs same data
// User copies entire result
After (efficient):
// Tool 1
{"results": [{"id": "r1", "name": "...", "preview": "10 lines"}]}
// Tool 2
input: {"id": "r1"} // Reference only
Before:
{
"results": [
{"name": "...", "full": "... 500 tokens ..."},
{"name": "...", "full": "... 500 tokens ..."},
{"name": "...", "full": "... 500 tokens ..."}
]
}
After:
{
"results": [
{"id": "a1", "conf": 0.95, "full": "..."}, // Only high confidence
{"id": "b2", "conf": 0.70, "summary": "..."},
{"id": "c3", "conf": 0.40} // Just ID
]
}
Before (15+ tools, no organization):
search_users, search_posts, get_user, get_post, ...
After (grouped):
Query Tools: search
Lookup Tools: get_user, get_post
Management: create_user, update_user
Without IDs:
With IDs:
Without aggregation:
With aggregation:
Proven patterns:
Key lessons:
Study these real-world examples when designing similar functionality.
npx claudepluginhub standardbeagle/standardbeagle-tools --plugin mcp-architectProvides design patterns for autonomous coding agents: agent loops, tool calling APIs, permissions, browser automation, human-in-the-loop workflows. Python examples included.
Provides MCP architecture patterns including client-host-server model, transports, resources, and tools with FastMCP examples in Python and TypeScript. Useful for building MCP servers and implementing tools.
Provides patterns for building secure MCP servers with OAuth auth, tool composition, elicitation, sampling, interactive UIs, and debugging. Use for MCP server development and integrations.