---

name: servers description: "Routes UltraRAG MCP server workflows, backend choices, and server-level tool and prompt signatures." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Servers

Use this sub-skill when the task is about the MCP servers under servers/.

Typical triggers

• servers/retriever, servers/generation, servers/prompt, servers/corpus, servers/evaluation, servers/custom, servers/memory, servers/benchmark, servers/router, or servers/sayhello
• Questions about tool or prompt signatures, outputs, or parameter keys
• Adding a new server tool or prompt
• Choosing between retriever backends, generation backends, or corpus/index backends
• Understanding generated server.yaml files or server build metadata
• Optional dependency failures inside a server import or backend initialization

What this sub-skill covers

• MCP server structure and the UltraRAG_MCP_Server wrapper.
• Verified tool/prompt signatures and output keys for the public server modules.
• Backend-specific configuration for retrieval, generation, corpus, evaluation, and web search.
• Source-level import problems such as missing optional dependencies or a stale package version.

What stays elsewhere

• Pipeline composition, example recipes, and ultrarag build/run usage belong in sub-skills/pipelines/.
• Flask routes, KB storage, and case-study behavior belong in sub-skills/ui-and-storage/.

Start here

• Read references/api-reference.md for the verified server signatures and output contracts.
• Read references/backends-and-config.md for backend choices, extras, and config keys.
• Read references/troubleshooting.md when a server import or backend choice fails.
• Run scripts/inspect_servers.py to print a live signature summary from a checkout-aware environment.

Common user questions this sub-skill should answer

• How do I add a tool to the retriever or generation server?
• Which parameters does retriever_search or generation.generate expect?
• Which backend is appropriate for embeddings, BM25, web search, or vLLM?
• Why does a server import fail on this machine?
• What optional package is needed for evaluate_trec, build_mineru_corpus, or show case?

Practical workflow

1. Identify the server family and the exact tool/prompt the task mentions.
2. Check the backend or extra that the server needs.
3. Confirm the registered input and output names before changing a pipeline.
4. Use the inspection script or the reference tables to validate the claim.

Helpful commands

Use the bundled inspection helper so checkout paths and local server imports are handled explicitly:

python sub-skills/servers/scripts/inspect_servers.py --repo-root <checkout>
python sub-skills/servers/scripts/inspect_servers.py --repo-root <checkout> --module servers.generation.src.generation

If the task is only about how to wire steps together, go back to the pipelines sub-skill instead of rewriting server guidance here.