The interface is localized. Skill descriptions and documents are machine-translated by AI into your language; code, commands, paths and technical terms stay in the original, and the author's document governs.
Explore skills
Development · SKILL.md

mcp-server

mcp-server

Guide strands-agents-mcp-server installation, CLI use, docs search/fetch semantics, llms.txt indexing, cache hydration, URL restrictions, environment variables, and offline/live tests.

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill documentation

GitHub

name: mcp-server description: "Guide strands-agents-mcp-server installation, CLI use, docs search/fetch semantics, llms.txt indexing, cache hydration, URL restrictions, environment variables, and offline/live tests." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

MCP server

Use this sub-skill when working on the strands-agents-mcp-server package and its documentation-access tools.

Start here

  1. Read overview.md for package boundaries, verified facts, and owned behavior.
  2. Read cli-reference.md for install, launch, environment variables, and safe check commands.
  3. Read workflows.md for search_docs, fetch_doc, hydration, sectioning, and test workflows.
  4. Read troubleshooting.md for expected error dictionaries, network issues, prefetch timing, and dependency warnings.
  5. Use mcp-smoke.sh for import/signature and console-help checks without network fetches.
  6. Use mcp-unit-check.sh for selected offline unit tests when invoked from a compatible checkout.

Use this sub-skill for

  • Installing or launching the docs-search MCP server through the console entry point, module entry point, or MCP client command.
  • search_docs(query, k=5) ranking, snippets, result fields, and hydration-sensitive body-term search.
  • fetch_doc(uri="", section="") catalog mode, TOC mode, section mode, small-document behavior, and error shapes.
  • llms.txt catalog indexing, cache hydration, snippet generation, background prefetch, retryable indexing failures, and section extraction.
  • URL restrictions, server environment variables, dependency warnings, offline unit tests, and live-doc integration tests.

Route elsewhere

  • Python SDK MCP client internals, transport adapters, or MCPClient usage by agents: use python-sdk.
  • Docs-site MDX pages, sourceLinks, generated API docs, or site navigation: use docs-site.
  • TypeScript SDK MCP client/tools: use typescript-sdk.

Operating rules

  • Do not fetch live documentation, run networked integration tests, or start long-lived inspectors by default.
  • Preserve the strict allowed host: only HTTPS URLs under strandsagents.com are valid document URLs.
  • Treat search scores as within-query ordering values, not probabilities or cross-query comparable numbers.
  • Keep unsupported URL, fetch failure, and unknown section errors as explicit dictionary returns unless the package contract changes deliberately.
  • Keep live-doc network tests separate from offline unit tests; skip or document them unless the task explicitly selects network verification.

Fast path

  1. Identify whether the task is packaging/launch, search ranking, fetch/sectioning, cache/indexing, URL validation, or tests.
  2. Read the matching reference and run the smoke helper in an environment where the package is installed.
  3. For behavior changes, prefer offline unit tests first; escalate to live docs only when the task requires real public site behavior.
  4. If a task crosses into SDK MCP client behavior, hand off to the Python or TypeScript SDK sub-skill instead of mixing contracts.
Saved skills