Inspired by: https://github.com/awslabs/mcp/tree/main/src/aws-documentation-mcp-server

OCI Documentation MCP Server

Model Context Protocol (MCP) server for OCI Documentation

This MCP server provides tools to search for content, and access OCI documentation.

Change log

• 2026-05-20: support transport: stdio,sse,streamable-http
• 2026-05-19: change search engine to oracle help center search
• 2025-04-21: Initial release

Features

oci_search_documentation

Searches OCI documentation through the Oracle Help Center Search API and returns structured page results. This tool is intended for the first step of a documentation workflow: finding the most relevant Oracle documentation URL before reading the page.

Parameters:

• search_phrase: Search text. Use specific OCI service names, product terms, error messages, or feature names for better results.
• limit: Maximum number of results to return. Defaults to 3.
• page: Search result page number. Defaults to 1.

Returns:

• Pagination metadata from the Oracle Help Center result set.
• A list of documentation results with title, URL, and description.

Design notes:

• Uses the public Oracle Help Center pages endpoint.

oci_read_documentation

Reads one OCI documentation page, converts it from HTML to Markdown, indexes it by line number, and returns a window of content. This tool is intended for controlled reading of long documentation pages without flooding the MCP response.

Parameters:

• url: OCI documentation page URL. The URL must be from docs.oracle.com and must end with .htm or .html.
• start_index: 0-based line number to start reading from. Defaults to 0.
• max_lines: Maximum number of Markdown lines to return. Defaults to 10.

Returns:

• stats: Total lines, total words, start line, returned lines, remaining lines, and remaining words.
• content: Markdown text for the requested line window.
• table_of_contents: Returned only when start_index == 0; includes heading level, title, and 0-based line number.

Design notes:

• Long documents are paged by Markdown line number rather than character offset, which makes follow-up reads easier for agents.
• Converted pages are cached in process memory for 24 hours, up to 128 pages. The cache stores a single line-list representation to avoid duplicating full Markdown text and split lines.
• Table of contents and related links are returned only for the first read to avoid repeating metadata during follow-up reads.

Use

Option 1: Run from pypi package

Defalt output through stdio, change that

{
  "mcpServers": {
    "oci-documentation-mcp-server": {
      "command": "uvx",
      "args": [
        "--from",
        "oci-documentation-mcp-server@latest",
        "python",
        "-m",
        "oci_documentation_mcp_server.server",
        "--transport",
        "stdio"
      ],
      "env": {
        "FASTMCP_LOG_LEVEL": "ERROR"
      }
    }
  }
}

Option 2: Run locally from source code and output through stdio

Installation Requirements

1. Install uv from Astral or the GitHub README
2. Install Python 3.10 or newer using uv python install 3.10 (or a more recent version)

{
  "mcpServers": {
    "oci-documentation-mcp-server": {
      "command": "uv",
      "args": [
        "--directory",
        "path/to/oci-documentation-mcp-server"
        "run",
        "python",
        "-m",
        "oci_documentation_mcp_server.server",
        "--transport",
        "stdio"
      ],
      "env": {
        "FASTMCP_LOG_LEVEL": "ERROR"
      }
    }
  }
}

Option 2: Run as server

Run the server use Streamable HTTP:

uv run python -m oci_documentation_mcp_server.server --transport "streamable-http" --port 8000 --path "/mcp"

Config on agent tools:

{
  "mcpServers": {
      "oci-documentation-mcp-server": {
      "type": "streamable-http",
      "url": "http://localhost:8000/mcp"
    }
  }
}

Tools

oci_search_documentation

Searches OCI documentation based on a search phrase.

oci_search_documentation(search_phrase: str, limit: int = 3, page: int = 1) -> str

oci_read_documentation

Fetches an OCI documentation page and returns content partially as markdown.

oci_read_documentation(url: str, start_index: int = 0, max_lines: int = 10) -> str