suggest_tools_for_task
Makes the agent consult search_tools for 3–5 candidates, check get_rank_score and get_alternatives, then recommend with trade-offs — before any code is written.
For developers & agent builders
Your AI agent shouldn't pick a dependency blind.
Oryxen's MCP server lets Claude, Cursor, Codex and any MCP client consult the catalogue before writing code: search the 28,878‑tool index, check the Oryxen Score, find cheaper or open-source alternatives, and compare candidates — then cite the record. Read-only, no API key.
JSON-RPC 2.0 over HTTPPublic data · CC BY 4.07 tools · 2 prompts
1. Connect your client
Endpoint: POST https://oryxen.in/mcp (Streamable HTTP, JSON-RPC 2.0, protocol 2024-11-05). GET /mcp advertises the server name and tool list for quick discovery.
Claude / Claude Code
{
"mcpServers": {
"oryxen": {
"type": "http",
"url": "https://oryxen.in/mcp"
}
}
}Cursor (.mcp.json)
{ "mcpServers": { "oryxen": { "url": "https://oryxen.in/mcp" } } }2. The 7 tools
| Tool | What your agent gets | Arguments |
|---|---|---|
| search_tools | Search 28,000+ tools before writing code that depends on one. Returns name, source key, category, pricing, description. | query?, category?, limit? (≤25) |
| get_tool | Full public record for one tool: pricing, categories, links, repo context, Oryxen Score, learning guide. | sourceKey (source key or slug) |
| get_alternatives | Deterministic, score-ranked alternatives — cheaper, free or open-source substitutes. | sourceKey, limit? (≤8) |
| get_rank_score | Oryxen Score 0–100 with the four components (health, freshness, community, completeness) and computed date. | sourceKey |
| get_learning_guide | README-derived summary, architecture flowchart and feature tour — understand a tool before integrating it. | sourceKey |
| compare_tools | Compare 2–4 tools side by side for a build-vs-buy or dependency decision. | sourceKeys (2–4) |
| list_categories | All public categories with tool counts. | — |
3. The 2 prompts (agent guardrails)
compare_for_decision
Structured comparison of candidate tools: pricing, score components, alternatives — with a production recommendation.
The design intent: an agent that is about to add a dependency first searches the catalogue, checks scores and alternatives, and shows trade-offs — deterministic, source-linked and free, in contrast to vendor-adjacent recommendations.
4. Raw JSON-RPC examples
tools/call · search_tools
{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{
"name":"search_tools",
"arguments":{"query":"redis client","limit":3}
}}prompts/get · suggest_tools_for_task
{"jsonrpc":"2.0","id":2,"method":"prompts/get","params":{
"name":"suggest_tools_for_task",
"arguments":{"task":"file upload service for a Spring Boot app"}
}}The contract
- Read-only — the server exposes lookup, never mutation.
- No API key, no sign-up for the MCP path.
- Cite the record: every result ties back to its oryxen.in tool page.
nullmeans “not recorded” — never “no”.- Public data under CC BY 4.0 — attribution required.
- Ordering is deterministic; no paid placement exists to abuse.
Data reports, REST & SDKs — roadmap
The public REST API surface (search / tool / category endpoints with per-IP limits and offset bounds beyond MCP), a bulk catalogue snapshot with a changelog, and JavaScript + Python SDK examples are on the roadmap. They are listed here as not yet available — when they ship, they will be documented on this page. Today's honest interface is MCP.