Add example servers for SEP-2053 variant patterns (#4)
Add four new example servers demonstrating different SEP-2053 variant patterns: - model-optimized (same tools, different descriptions per LLM family) - trading (API versioning with deprecation lifecycle) - research (context budget management) - Refactor the existing variants-http example into a GitHub-style server with custom ranking. Signed-off-by: Kurt Degiorgio <kdegiorgio@bloomberg.net>
This commit is contained in:
parent
c01fd188d2
commit
e1c17e047f
1 changed files with 1218 additions and 147 deletions
139
go/sdk/examples/server/research/main.go
Normal file
139
go/sdk/examples/server/research/main.go
Normal file
|
|
@ -0,0 +1,139 @@
|
|||
// Example: Research Assistant — context budget management with the same tools
|
||||
// at different verbosity levels. Demonstrates how variants can optimize tool
|
||||
// descriptions for different context window sizes and use cases.
|
||||
//
|
||||
// Capability demonstrated: Context budget management, description verbosity.
|
||||
//
|
||||
// Variants:
|
||||
// - deep-research: Verbose multi-paragraph descriptions with usage examples
|
||||
// - quick-lookup: Concise 1-sentence descriptions for fast Q&A
|
||||
// - synthesis: Balanced descriptions for report generation
|
||||
//
|
||||
// Run:
|
||||
//
|
||||
// go run ./examples/server/research
|
||||
//
|
||||
// Then connect any MCP client to http://localhost:8080.
|
||||
package main
|
||||
|
||||
import (
|
||||
"context"
|
||||
"log"
|
||||
"net/http"
|
||||
"slices"
|
||||
"strings"
|
||||
|
||||
"github.com/modelcontextprotocol/go-sdk/mcp"
|
||||
|
||||
"github.com/modelcontextprotocol/experimental-ext-variants/go/sdk/variants"
|
||||
)
|
||||
|
||||
func main() {
|
||||
// Deep-research variant: verbose descriptions with usage examples
|
||||
// for thorough research workflows with large context windows.
|
||||
deepServer := mcp.NewServer(&mcp.Implementation{Name: "research-assistant", Version: "v1.0.0"}, nil)
|
||||
mcp.AddTool(deepServer, &mcp.Tool{
|
||||
Name: "search_papers",
|
||||
Description: "Search academic papers across multiple databases including arXiv, Semantic Scholar, and PubMed. " +
|
||||
"Use this tool to find relevant literature for a research topic. Supports filtering by year and field. " +
|
||||
"Returns paper metadata including titles, authors, abstracts, and citation counts. " +
|
||||
"For best results, use specific technical terms rather than broad queries. " +
|
||||
"Example queries: 'transformer attention mechanisms in computer vision', " +
|
||||
"'reinforcement learning from human feedback alignment'. " +
|
||||
"Follow up with get_paper to retrieve full details for promising results.",
|
||||
}, searchPapers)
|
||||
mcp.AddTool(deepServer, &mcp.Tool{
|
||||
Name: "get_paper",
|
||||
Description: "Retrieve full details for a specific academic paper by its identifier. " +
|
||||
"Accepts arXiv IDs (e.g., arxiv:2301.01234), DOIs (e.g., doi:10.1234/5678), or Semantic Scholar IDs. " +
|
||||
"Returns complete metadata including abstract, references, keywords, and publication venue. " +
|
||||
"Use this after search_papers to get deeper context on papers of interest. " +
|
||||
"The references list can be used to find related work by calling get_paper recursively. " +
|
||||
"Combine with summarize to generate concise summaries of lengthy papers.",
|
||||
}, getPaper)
|
||||
mcp.AddTool(deepServer, &mcp.Tool{
|
||||
Name: "summarize",
|
||||
Description: "Generate a summary of the provided text using various styles and lengths. " +
|
||||
"Styles: 'abstract' produces a formal academic abstract, 'bullet-points' extracts key findings " +
|
||||
"as a structured list, 'executive' creates a high-level overview for non-technical audiences. " +
|
||||
"Lengths: 'short' (1-2 sentences), 'medium' (1 paragraph), 'long' (2-3 paragraphs). " +
|
||||
"Best used with paper abstracts or full text retrieved via get_paper. " +
|
||||
"For multi-paper synthesis, concatenate abstracts and use the 'executive' style.",
|
||||
}, summarize)
|
||||
|
||||
// Quick-lookup variant: concise descriptions for fast Q&A sessions
|
||||
// where context budget is limited.
|
||||
quickServer := mcp.NewServer(&mcp.Implementation{Name: "research-assistant", Version: "v1.0.0"}, nil)
|
||||
mcp.AddTool(quickServer, &mcp.Tool{
|
||||
Name: "search_papers",
|
||||
Description: "Search academic papers by query, with optional year and field filters.",
|
||||
}, searchPapers)
|
||||
mcp.AddTool(quickServer, &mcp.Tool{
|
||||
Name: "get_paper",
|
||||
Description: "Get full details for a paper by its ID (arXiv, DOI, etc.).",
|
||||
}, getPaper)
|
||||
mcp.AddTool(quickServer, &mcp.Tool{
|
||||
Name: "summarize",
|
||||
Description: "Summarize text in abstract, bullet-point, or executive style.",
|
||||
}, summarize)
|
||||
|
||||
// Synthesis variant: balanced descriptions for report generation
|
||||
// workflows that need moderate detail.
|
||||
synthesisServer := mcp.NewServer(&mcp.Implementation{Name: "research-assistant", Version: "v1.0.0"}, nil)
|
||||
mcp.AddTool(synthesisServer, &mcp.Tool{
|
||||
Name: "search_papers",
|
||||
Description: "Search academic papers across arXiv, Semantic Scholar, and PubMed. " +
|
||||
"Returns titles, authors, abstracts, and citation counts. Filter by year and field.",
|
||||
}, searchPapers)
|
||||
mcp.AddTool(synthesisServer, &mcp.Tool{
|
||||
Name: "get_paper",
|
||||
Description: "Retrieve full paper details including abstract, references, keywords, and venue. " +
|
||||
"Accepts arXiv IDs, DOIs, or Semantic Scholar IDs.",
|
||||
}, getPaper)
|
||||
mcp.AddTool(synthesisServer, &mcp.Tool{
|
||||
Name: "summarize",
|
||||
Description: "Generate a summary in abstract, bullet-point, or executive style. " +
|
||||
"Configurable length: short (1-2 sentences), medium (paragraph), or long (2-3 paragraphs).",
|
||||
}, summarize)
|
||||
|
||||
vs := variants.NewServer(&mcp.Implementation{Name: "research-assistant", Version: "v1.0.0"}).
|
||||
WithVariant(variants.ServerVariant{
|
||||
ID: "deep-research",
|
||||
Description: "Verbose tool descriptions with usage examples and guidance for thorough research workflows. Best for agents with large context windows performing literature reviews or deep analysis.",
|
||||
Hints: map[string]string{"contextSize": "verbose", "useCase": "research"},
|
||||
Status: variants.Stable,
|
||||
}, deepServer, 0).
|
||||
WithVariant(variants.ServerVariant{
|
||||
ID: "quick-lookup",
|
||||
Description: "Concise 1-sentence tool descriptions for fast question-answering. Minimal context usage for agents with limited token budgets or simple lookup tasks.",
|
||||
Hints: map[string]string{"contextSize": "compact", "useCase": "qa"},
|
||||
Status: variants.Stable,
|
||||
}, quickServer, 1).
|
||||
WithVariant(variants.ServerVariant{
|
||||
ID: "synthesis",
|
||||
Description: "Balanced tool descriptions for report generation and multi-paper synthesis. Moderate detail level suitable for structured writing workflows.",
|
||||
Hints: map[string]string{"contextSize": "standard", "useCase": "synthesis"},
|
||||
Status: variants.Experimental,
|
||||
}, synthesisServer, 2).
|
||||
// Custom ranking: match by contextSize hint, fall back to priority.
|
||||
WithRanking(func(_ context.Context, hints variants.VariantHints, vs []variants.ServerVariant) []variants.ServerVariant {
|
||||
requested, _ := variants.HintValue[string](hints, "contextSize")
|
||||
slices.SortStableFunc(vs, func(a, b variants.ServerVariant) int {
|
||||
aMatch := strings.EqualFold(a.Hints["contextSize"], requested)
|
||||
bMatch := strings.EqualFold(b.Hints["contextSize"], requested)
|
||||
if aMatch != bMatch {
|
||||
if aMatch {
|
||||
return -1
|
||||
}
|
||||
return 1
|
||||
}
|
||||
return a.Priority() - b.Priority()
|
||||
})
|
||||
return vs
|
||||
})
|
||||
|
||||
handler := variants.NewStreamableHTTPHandler(vs, nil)
|
||||
|
||||
log.Println("Listening on :8080")
|
||||
log.Fatal(http.ListenAndServe(":8080", handler))
|
||||
}
|
||||
Loading…
Reference in a new issue