• Docker installed and running

• uv

  installed (for running Python scripts)
• The

  opensearch-launchpad

  repository cloned locally

{
  "mcpServers": {
    "ddg-search": {
      "command": "uvx",
      "args": ["duckduckgo-mcp-server"]
    },
    "awslabs.aws-api-mcp-server": {
      "command": "uvx",
      "args": ["awslabs.aws-api-mcp-server@latest"],
      "env": { "FASTMCP_LOG_LEVEL": "ERROR" }
    },
    "aws-knowledge-mcp-server": {
      "command": "uvx",
      "args": ["fastmcp", "run", "https://knowledge-mcp.global.api.aws"],
      "env": { "FASTMCP_LOG_LEVEL": "ERROR" }
    },
    "opensearch-mcp-server": {
      "command": "uvx",
      "args": ["opensearch-mcp-server-py@latest"],
      "env": { "FASTMCP_LOG_LEVEL": "ERROR" }
    }
  }
}

• ddg-search

  — Search OpenSearch documentation via DuckDuckGo. Use

  search(query="site:opensearch.org <your query>")

  to find docs, then

  fetch_content(url)

  to read the full page.

• awslabs.aws-api-mcp-server

  — AWS API calls (required for Phase 5 deployment, also useful for general AWS questions).

• aws-knowledge-mcp-server

  — AWS documentation lookup.

• opensearch-mcp-server

  — Direct OpenSearch API access on local and remote clusters, including Amazon OpenSearch Service (AOS) and Serverless (AOSS). Handles SigV4 auth transparently. See the User Guide for full configuration options.

1. Locate the MCP config file for the current IDE:
   • Kiro:

     .kiro/settings/mcp.json

   • Cursor:

     .cursor/mcp.json

   • Claude Code:

     .mcp.json

   • VS Code (Copilot):

     .vscode/mcp.json

   • Windsurf:

     ~/.codeium/windsurf/mcp_config.json

   • If unsure, check for any of the above files in the workspace root.
2. Read the existing config (or start with

   {"mcpServers": {}}

   if the file doesn't exist).
3. Merge in the missing server entry from the JSON block above. Do not overwrite existing entries.
4. Save the file.
5. Inform the user: "I've added the [server name] MCP server to your config. Please restart your IDE or reconnect MCP servers for the changes to take effect."
6. Wait for the user to confirm the restart, then retry the tool call.

1. Run

   opensearch_ops.py search-docs --query "<your query>"

   to search opensearch.org (default).
   • Covers

     docs.opensearch.org

     (APIs, configuration, query DSL, plugins),

     opensearch.org/blog

     (release announcements, feature deep-dives), and

     opensearch.org/platform

     .
2. For AWS-specific questions (e.g. Amazon OpenSearch Service, Serverless, IAM policies, pricing), use

   --site docs.aws.amazon.com

   :
   bash

   uv run python scripts/opensearch_ops.py search-docs --query "OpenSearch Serverless pricing" --site docs.aws.amazon.com

3. Review the returned titles, URLs, and snippets.
4. If more detail is needed, fetch the full page content from the top result URL.
5. Summarize the answer based on the documentation.

uv run python scripts/opensearch_ops.py search-docs --query "OpenSearch 3.5 features"
uv run python scripts/opensearch_ops.py search-docs --query "neural sparse search" --count 3
uv run python scripts/opensearch_ops.py search-docs --query "OpenSearch Service domain access policy" --site docs.aws.amazon.com

scripts/

• start_opensearch.sh

  — Start a local OpenSearch cluster via Docker

• opensearch_ops.py

  — CLI for all OpenSearch operations. See CLI Reference for exact invocations and examples

bash scripts/start_opensearch.sh
uv run python scripts/opensearch_ops.py <command> [options]

• Ask one preference question per message.
• Never skip Phase 1 (sample document collection).
• Show architecture proposals to the user before execution.
• Follow the phases in order — do not jump ahead.
• When a step fails, present the error and wait for guidance.

uv run python scripts/opensearch_ops.py preflight-check

• status: "available"

  — A cluster is already running and reachable. Use it directly. The

  auth_mode

  field shows which authentication was detected (

  none

  ,

  default

  , or

  custom

  ).

• status: "auth_required"

  — A cluster is running but requires credentials. Ask the user for their username and password, then run:
  bash

  uv run python scripts/opensearch_ops.py preflight-check --auth-mode custom --username <user> --password <pass>

  If successful, the credentials are persisted for the session and all subsequent commands will use them automatically.

• status: "no_cluster"

  — No cluster detected. Start one:
  bash

  bash scripts/start_opensearch.sh

load-sample

references/knowledge/document_processing_guide.md

1. Install Docling:

   uv pip install docling

2. Convert the document using Docling's

   DocumentConverter

   to extract structured text.
3. Chunk the output using

   HybridChunker(max_tokens=512, overlap_tokens=50)

   .
4. Export chunks as JSONL and use

   opensearch_ops.py index-bulk

   to load them.

bm25

dense_vector

neural_sparse

hybrid

agentic

• references/knowledge/dense_vector_models.md

• references/knowledge/sparse_vector_models.md

• references/knowledge/opensearch_semantic_search_guide.md

• references/knowledge/agentic_search_guide.md

• references/knowledge/document_processing_guide.md

  (when working with PDF/DOCX/PPTX sources)

opensearch_ops.py

opensearch_ops.py --help

http://127.0.0.1:8765

"Your search app is live! Here's what you can do next:"

1. Evaluate search quality (Phase 4.5) — I'll run test queries, measure relevance metrics (nDCG, precision, MRR), and suggest improvements.
2. Deploy to Amazon OpenSearch Service (Phase 5) — Provision an Amazon OpenSearch cluster and deploy your search setup.
3. Done for now — Keep experimenting with the Search Builder UI.