openrag-mcp 0.1.0__tar.gz

openrag_mcp-0.1.0/.gitignore

@@ -0,0 +1,44 @@
+# Python-generated files
+__pycache__/
+*.py[oc]
+build/
+dist/
+wheels/
+*.egg-info
+
+# Virtual environments
+.venv
+.env
+
+# RSA keys for JWT signing
+*.pem
+
+.idea/
+
+1001*.pdf
+*.json
+!**/package.json
+!**/package-lock.json
+!**/tsconfig.json
+!flows/*.json
+!src/tui/_assets/flows/*.json
+!src/tui/_assets/flows/components/*.json
+!frontend/*.json
+.DS_Store
+
+/config/
+
+.docling.pid
+
+# OpenSearch data directory
+opensearch-data/
+
+node_modules
+
+# AI Tools
+/.claude
+/CLAUDE.md
+/.gemini
+/GEMINI.md
+/.bob
+/AGENTS.md

openrag_mcp-0.1.0/PKG-INFO

@@ -0,0 +1,257 @@
+Metadata-Version: 2.4
+Name: openrag-mcp
+Version: 0.1.0
+Summary: MCP server for OpenRAG
+Author: OpenRAG Team
+License: Apache-2.0
+Keywords: ai,llm,mcp,openrag,rag
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Requires-Python: >=3.10
+Requires-Dist: httpx>=0.27.0
+Requires-Dist: mcp>=1.0.0
+Requires-Dist: openrag-sdk>=0.1.0
+Description-Content-Type: text/markdown
+
+# OpenRAG MCP Server
+
+An [Model Context Protocol](https://modelcontextprotocol.io/) (MCP) server that exposes your OpenRAG knowledge base to AI assistants. It lets MCP-compatible apps like Cursor, Claude Desktop, and IBM Watson Orchestrate use OpenRAG’s RAG capabilities (chat, search, settings) over a standard protocol—no custom integrations per platform.
+
+---
+
+## What is OpenRAG MCP?
+
+OpenRAG MCP is a **connectivity layer** between your OpenRAG instance and AI applications. The host app (e.g. Cursor or Claude Desktop) runs the MCP server as a subprocess and talks to it over stdio using JSON-RPC. The server then calls your OpenRAG API with your API key. Your knowledge base stays the single source of truth; all connected apps get the same RAG-backed chat and search.
+
+---
+
+## Quick Start
+
+Run the server with **uvx** (no local install required; requires Python 3.10+ and [uv](https://docs.astral.sh/uv/)):
+
+```bash
+uvx openrag-mcp
+```
+
+Set required environment variables first (or pass them via your MCP client config):
+
+```bash
+export OPENRAG_URL="https://your-openrag-instance.com"
+export OPENRAG_API_KEY="orag_your_api_key"
+uvx openrag-mcp
+```
+
+To pin a version:
+
+```bash
+uvx --from openrag-mcp==0.2.1 openrag-mcp
+```
+
+### Prerequisites
+
+- Python 3.10+
+- A running OpenRAG instance
+- An OpenRAG API key (create one in **Settings → API Keys** in OpenRAG)
+- `uv` installed (for `uvx`)
+
+---
+
+## Available Tools
+
+These tools are currently exposed by the server:
+
+| Tool | Description |
+|:-----|:------------|
+| `openrag_chat` | Send a message and get a RAG-enhanced response. Optional: `chat_id`, `filter_id`, `limit`, `score_threshold`. |
+| `openrag_search` | Semantic search over the knowledge base. Optional: `limit`, `score_threshold`, `filter_id`, `data_sources`, `document_types`. |
+| `openrag_get_settings` | Get current OpenRAG configuration (LLM, embeddings, chunk settings, system prompt, etc.). |
+| `openrag_update_settings` | Update OpenRAG configuration (LLM model, embedding model, chunk size/overlap, system prompt, table structure, OCR, picture descriptions). |
+| `openrag_list_models` | List available language and embedding models for a provider (`openai`, `anthropic`, `ollama`, `watsonx`). |
+
+### Coming later (document tools)
+
+Document ingestion and management tools (`openrag_ingest_file`, `openrag_ingest_url`, `openrag_delete_document`, `openrag_get_task_status`, `openrag_wait_for_task`) are implemented but not yet registered in this server; they will be enabled in a future release.
+
+---
+
+## Environment Variables
+
+| Variable | Description | Required | Default |
+|:---------|:------------|:--------:|:--------|
+| `OPENRAG_API_KEY` | Your OpenRAG API key | Yes | — |
+| `OPENRAG_URL` | Base URL of your OpenRAG instance | No | `http://localhost:3000` |
+
+**MCP HTTP client (optional):**
+
+| Variable | Description | Required | Default |
+|:---------|:------------|:--------:|:--------|
+| `OPENRAG_MCP_TIMEOUT` | Request timeout in seconds | No | `60.0` |
+| `OPENRAG_MCP_MAX_CONNECTIONS` | Maximum concurrent connections | No | `100` |
+| `OPENRAG_MCP_MAX_KEEPALIVE_CONNECTIONS` | Maximum keepalive connections | No | `20` |
+| `OPENRAG_MCP_MAX_RETRIES` | Maximum retry attempts for failed requests | No | `3` |
+| `OPENRAG_MCP_FOLLOW_REDIRECTS` | Whether to follow HTTP redirects | No | `true` |
+
+These must be set in the environment when the MCP server runs (e.g. in the `env` block of your MCP client config).
+
+---
+
+## How to Use
+
+### Cursor
+
+**Config file:** `~/.cursor/mcp.json`
+
+```json
+{
+  "mcpServers": {
+    "openrag": {
+      "command": "uvx",
+      "args": ["openrag-mcp"],
+      "env": {
+        "OPENRAG_URL": "https://your-openrag-instance.com",
+        "OPENRAG_API_KEY": "orag_your_api_key_here"
+      }
+    }
+  }
+}
+```
+
+Restart Cursor after changing the config.
+
+### Claude Desktop
+
+**macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json`  
+**Windows:** `%APPDATA%\Claude\claude_desktop_config.json`
+
+```json
+{
+  "mcpServers": {
+    "openrag": {
+      "command": "uvx",
+      "args": ["openrag-mcp"],
+      "env": {
+        "OPENRAG_URL": "https://your-openrag-instance.com",
+        "OPENRAG_API_KEY": "orag_your_api_key_here"
+      }
+    }
+  }
+}
+```
+
+Restart Claude Desktop after editing the file.
+
+---
+
+## Run from source (development)
+
+To use the **latest MCP code** from the repo (including settings and models tools), run from source. Do **not** install the package if you want local edits to apply.
+
+### Steps
+
+| Step | What | Command | Required for |
+|------|------|---------|---------------|
+| 1 | OpenRAG backend | Run your OpenRAG app (e.g. frontend + API) | All tools |
+| 2 | MCP from source | `cd sdks/mcp && uv sync` | All tools; no wheel needed |
+| 3 | (Optional) SDK from repo | `cd sdks/python && uv pip install -e .` | Only if you need unreleased chat/search SDK changes |
+
+Settings and models tools (`openrag_get_settings`, `openrag_update_settings`, `openrag_list_models`) use direct HTTP. Chat and search use the OpenRAG SDK (PyPI version is fine unless you need unreleased SDK changes).
+
+### Run the MCP from source
+
+```bash
+cd sdks/mcp
+uv sync
+export OPENRAG_URL="http://localhost:3000"
+export OPENRAG_API_KEY="orag_your_api_key"
+uv run openrag-mcp
+```
+
+### Cursor: use repo path so it runs your code
+
+In `~/.cursor/mcp.json`, set `--directory` to your **actual repo path** so Cursor runs the MCP from source:
+
+```json
+{
+  "mcpServers": {
+    "openrag": {
+      "command": "uv",
+      "args": [
+        "run",
+        "--directory",
+        "/path/to/openrag/sdks/mcp",
+        "openrag-mcp"
+      ],
+      "env": {
+        "OPENRAG_URL": "https://your-openrag-instance.com",
+        "OPENRAG_API_KEY": "orag_your_api_key_here"
+      }
+    }
+  }
+}
+```
+
+Replace `/path/to/openrag` with your real path (e.g. `/Users/edwin.jose/Documents/openrag`).
+
+If you previously installed the MCP (`pip install openrag-mcp` or a wheel), uninstall it so Cursor uses the repo:
+
+```bash
+uv pip uninstall openrag-mcp
+```
+
+Then restart Cursor.
+
+---
+
+## Use cases and benefits
+
+- **One integration, many apps** – Same MCP server works with Cursor, Claude Desktop, Watson Orchestrate, and any MCP client.
+- **RAG in the loop** – Chat and search are grounded in your OpenRAG knowledge base, with optional filters and scoring.
+- **Agent-friendly** – Agents can call OpenRAG for answers, list models, and read/update settings without custom APIs.
+- **Lightweight** – No extra service to deploy; the host app spawns the server as a subprocess and talks over stdio.
+- **Secure** – Only clients that have your `OPENRAG_API_KEY` (via env) can use the server to access OpenRAG.
+
+**Example scenarios:** Query internal docs and runbooks from your IDE; power support bots with your product docs; search and summarize across ingested documents; automate workflows that need RAG (when document tools are enabled).
+
+---
+
+
+## Example prompts
+
+Once the server is configured, you can ask the AI to:
+
+- *"Search my knowledge base for authentication best practices"*
+- *"Chat with OpenRAG about the Q4 roadmap"*
+- *"What are the current OpenRAG settings?"*
+- *"List available models for the openai provider"*
+- *"Update OpenRAG to use chunk size 512"*
+
+---
+
+## Troubleshooting
+
+### "OPENRAG_API_KEY environment variable is required"
+
+Set `OPENRAG_API_KEY` in the `env` section of your MCP config (Cursor or Claude Desktop). The server reads it at startup.
+
+### "Connection refused" or network errors
+
+1. Confirm your OpenRAG instance is running and reachable.
+2. Check `OPENRAG_URL` (no trailing slash; include `https://` if applicable).
+3. Ensure no firewall or proxy is blocking the client machine from reaching OpenRAG.
+
+### Tools not appearing
+
+1. Restart the host app (Cursor or Claude Desktop) after changing the MCP config.
+2. Check the app’s MCP/log output for errors (e.g. wrong `command`/`args` or missing `uv`/`uvx`).
+3. If using "run from source", ensure `args` includes `--directory` and the correct path to `sdks/mcp`.
+
+---
+
+## License
+
+Apache 2.0 - See [LICENSE](../../LICENSE) for details.

openrag_mcp-0.1.0/README.md

@@ -0,0 +1,237 @@
+# OpenRAG MCP Server
+
+An [Model Context Protocol](https://modelcontextprotocol.io/) (MCP) server that exposes your OpenRAG knowledge base to AI assistants. It lets MCP-compatible apps like Cursor, Claude Desktop, and IBM Watson Orchestrate use OpenRAG’s RAG capabilities (chat, search, settings) over a standard protocol—no custom integrations per platform.
+
+---
+
+## What is OpenRAG MCP?
+
+OpenRAG MCP is a **connectivity layer** between your OpenRAG instance and AI applications. The host app (e.g. Cursor or Claude Desktop) runs the MCP server as a subprocess and talks to it over stdio using JSON-RPC. The server then calls your OpenRAG API with your API key. Your knowledge base stays the single source of truth; all connected apps get the same RAG-backed chat and search.
+
+---
+
+## Quick Start
+
+Run the server with **uvx** (no local install required; requires Python 3.10+ and [uv](https://docs.astral.sh/uv/)):
+
+```bash
+uvx openrag-mcp
+```
+
+Set required environment variables first (or pass them via your MCP client config):
+
+```bash
+export OPENRAG_URL="https://your-openrag-instance.com"
+export OPENRAG_API_KEY="orag_your_api_key"
+uvx openrag-mcp
+```
+
+To pin a version:
+
+```bash
+uvx --from openrag-mcp==0.2.1 openrag-mcp
+```
+
+### Prerequisites
+
+- Python 3.10+
+- A running OpenRAG instance
+- An OpenRAG API key (create one in **Settings → API Keys** in OpenRAG)
+- `uv` installed (for `uvx`)
+
+---
+
+## Available Tools
+
+These tools are currently exposed by the server:
+
+| Tool | Description |
+|:-----|:------------|
+| `openrag_chat` | Send a message and get a RAG-enhanced response. Optional: `chat_id`, `filter_id`, `limit`, `score_threshold`. |
+| `openrag_search` | Semantic search over the knowledge base. Optional: `limit`, `score_threshold`, `filter_id`, `data_sources`, `document_types`. |
+| `openrag_get_settings` | Get current OpenRAG configuration (LLM, embeddings, chunk settings, system prompt, etc.). |
+| `openrag_update_settings` | Update OpenRAG configuration (LLM model, embedding model, chunk size/overlap, system prompt, table structure, OCR, picture descriptions). |
+| `openrag_list_models` | List available language and embedding models for a provider (`openai`, `anthropic`, `ollama`, `watsonx`). |
+
+### Coming later (document tools)
+
+Document ingestion and management tools (`openrag_ingest_file`, `openrag_ingest_url`, `openrag_delete_document`, `openrag_get_task_status`, `openrag_wait_for_task`) are implemented but not yet registered in this server; they will be enabled in a future release.
+
+---
+
+## Environment Variables
+
+| Variable | Description | Required | Default |
+|:---------|:------------|:--------:|:--------|
+| `OPENRAG_API_KEY` | Your OpenRAG API key | Yes | — |
+| `OPENRAG_URL` | Base URL of your OpenRAG instance | No | `http://localhost:3000` |
+
+**MCP HTTP client (optional):**
+
+| Variable | Description | Required | Default |
+|:---------|:------------|:--------:|:--------|
+| `OPENRAG_MCP_TIMEOUT` | Request timeout in seconds | No | `60.0` |
+| `OPENRAG_MCP_MAX_CONNECTIONS` | Maximum concurrent connections | No | `100` |
+| `OPENRAG_MCP_MAX_KEEPALIVE_CONNECTIONS` | Maximum keepalive connections | No | `20` |
+| `OPENRAG_MCP_MAX_RETRIES` | Maximum retry attempts for failed requests | No | `3` |
+| `OPENRAG_MCP_FOLLOW_REDIRECTS` | Whether to follow HTTP redirects | No | `true` |
+
+These must be set in the environment when the MCP server runs (e.g. in the `env` block of your MCP client config).
+
+---
+
+## How to Use
+
+### Cursor
+
+**Config file:** `~/.cursor/mcp.json`
+
+```json
+{
+  "mcpServers": {
+    "openrag": {
+      "command": "uvx",
+      "args": ["openrag-mcp"],
+      "env": {
+        "OPENRAG_URL": "https://your-openrag-instance.com",
+        "OPENRAG_API_KEY": "orag_your_api_key_here"
+      }
+    }
+  }
+}
+```
+
+Restart Cursor after changing the config.
+
+### Claude Desktop
+
+**macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json`  
+**Windows:** `%APPDATA%\Claude\claude_desktop_config.json`
+
+```json
+{
+  "mcpServers": {
+    "openrag": {
+      "command": "uvx",
+      "args": ["openrag-mcp"],
+      "env": {
+        "OPENRAG_URL": "https://your-openrag-instance.com",
+        "OPENRAG_API_KEY": "orag_your_api_key_here"
+      }
+    }
+  }
+}
+```
+
+Restart Claude Desktop after editing the file.
+
+---
+
+## Run from source (development)
+
+To use the **latest MCP code** from the repo (including settings and models tools), run from source. Do **not** install the package if you want local edits to apply.
+
+### Steps
+
+| Step | What | Command | Required for |
+|------|------|---------|---------------|
+| 1 | OpenRAG backend | Run your OpenRAG app (e.g. frontend + API) | All tools |
+| 2 | MCP from source | `cd sdks/mcp && uv sync` | All tools; no wheel needed |
+| 3 | (Optional) SDK from repo | `cd sdks/python && uv pip install -e .` | Only if you need unreleased chat/search SDK changes |
+
+Settings and models tools (`openrag_get_settings`, `openrag_update_settings`, `openrag_list_models`) use direct HTTP. Chat and search use the OpenRAG SDK (PyPI version is fine unless you need unreleased SDK changes).
+
+### Run the MCP from source
+
+```bash
+cd sdks/mcp
+uv sync
+export OPENRAG_URL="http://localhost:3000"
+export OPENRAG_API_KEY="orag_your_api_key"
+uv run openrag-mcp
+```
+
+### Cursor: use repo path so it runs your code
+
+In `~/.cursor/mcp.json`, set `--directory` to your **actual repo path** so Cursor runs the MCP from source:
+
+```json
+{
+  "mcpServers": {
+    "openrag": {
+      "command": "uv",
+      "args": [
+        "run",
+        "--directory",
+        "/path/to/openrag/sdks/mcp",
+        "openrag-mcp"
+      ],
+      "env": {
+        "OPENRAG_URL": "https://your-openrag-instance.com",
+        "OPENRAG_API_KEY": "orag_your_api_key_here"
+      }
+    }
+  }
+}
+```
+
+Replace `/path/to/openrag` with your real path (e.g. `/Users/edwin.jose/Documents/openrag`).
+
+If you previously installed the MCP (`pip install openrag-mcp` or a wheel), uninstall it so Cursor uses the repo:
+
+```bash
+uv pip uninstall openrag-mcp
+```
+
+Then restart Cursor.
+
+---
+
+## Use cases and benefits
+
+- **One integration, many apps** – Same MCP server works with Cursor, Claude Desktop, Watson Orchestrate, and any MCP client.
+- **RAG in the loop** – Chat and search are grounded in your OpenRAG knowledge base, with optional filters and scoring.
+- **Agent-friendly** – Agents can call OpenRAG for answers, list models, and read/update settings without custom APIs.
+- **Lightweight** – No extra service to deploy; the host app spawns the server as a subprocess and talks over stdio.
+- **Secure** – Only clients that have your `OPENRAG_API_KEY` (via env) can use the server to access OpenRAG.
+
+**Example scenarios:** Query internal docs and runbooks from your IDE; power support bots with your product docs; search and summarize across ingested documents; automate workflows that need RAG (when document tools are enabled).
+
+---
+
+
+## Example prompts
+
+Once the server is configured, you can ask the AI to:
+
+- *"Search my knowledge base for authentication best practices"*
+- *"Chat with OpenRAG about the Q4 roadmap"*
+- *"What are the current OpenRAG settings?"*
+- *"List available models for the openai provider"*
+- *"Update OpenRAG to use chunk size 512"*
+
+---
+
+## Troubleshooting
+
+### "OPENRAG_API_KEY environment variable is required"
+
+Set `OPENRAG_API_KEY` in the `env` section of your MCP config (Cursor or Claude Desktop). The server reads it at startup.
+
+### "Connection refused" or network errors
+
+1. Confirm your OpenRAG instance is running and reachable.
+2. Check `OPENRAG_URL` (no trailing slash; include `https://` if applicable).
+3. Ensure no firewall or proxy is blocking the client machine from reaching OpenRAG.
+
+### Tools not appearing
+
+1. Restart the host app (Cursor or Claude Desktop) after changing the MCP config.
+2. Check the app’s MCP/log output for errors (e.g. wrong `command`/`args` or missing `uv`/`uvx`).
+3. If using "run from source", ensure `args` includes `--directory` and the correct path to `sdks/mcp`.
+
+---
+
+## License
+
+Apache 2.0 - See [LICENSE](../../LICENSE) for details.

openrag_mcp-0.1.0/pyproject.toml

@@ -0,0 +1,34 @@
+[project]
+name = "openrag-mcp"
+version = "0.1.0"
+description = "MCP server for OpenRAG"
+readme = "README.md"
+requires-python = ">=3.10"
+license = { text = "Apache-2.0" }
+authors = [{ name = "OpenRAG Team" }]
+keywords = ["mcp", "openrag", "rag", "ai", "llm"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: Apache Software License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+]
+dependencies = [
+    "mcp>=1.0.0",
+    "httpx>=0.27.0",
+    "openrag-sdk>=0.1.0",
+]
+
+[project.scripts]
+openrag-mcp = "openrag_mcp:main"
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/openrag_mcp"]
+

openrag_mcp-0.1.0/src/openrag_mcp/__init__.py

@@ -0,0 +1,7 @@
+"""OpenRAG MCP Server - Expose OpenRAG capabilities via Model Context Protocol."""
+
+from openrag_mcp.server import main
+
+__version__ = "0.1.0"
+__all__ = ["main"]
+

openrag_mcp-0.1.0/src/openrag_mcp/__main__.py

@@ -0,0 +1,7 @@
+"""Entry point for running as a module: python -m openrag_mcp"""
+
+from openrag_mcp.server import main
+
+if __name__ == "__main__":
+    main()
+

openrag_mcp-0.1.0/src/openrag_mcp/config.py

@@ -0,0 +1,111 @@
+"""Configuration for OpenRAG MCP server."""
+
+import os
+
+from openrag_sdk import OpenRAGClient
+
+
+def _parse_float(key: str, default: float) -> float:
+    """Parse a positive float from environment."""
+    raw = os.environ.get(key)
+    if raw is None:
+        return default
+    try:
+        value = float(raw)
+        return value if value > 0 else default
+    except ValueError:
+        return default
+
+
+def _parse_int(key: str, default: int) -> int:
+    """Parse a positive int from environment."""
+    raw = os.environ.get(key)
+    if raw is None:
+        return default
+    try:
+        value = int(raw)
+        return value if value > 0 else default
+    except ValueError:
+        return default
+
+
+def _parse_bool(key: str, default: bool) -> bool:
+    """Parse a boolean from environment (true/false, 1/0)."""
+    raw = os.environ.get(key)
+    if raw is None:
+        return default
+    return raw.strip().lower() in ("true", "1", "yes")
+
+
+class Config:
+    """Configuration loaded from environment variables."""
+
+    def __init__(self):
+        self.openrag_url = os.environ.get("OPENRAG_URL", "http://localhost:3000")
+        self.api_key = os.environ.get("OPENRAG_API_KEY")
+
+        if not self.api_key:
+            raise ValueError(
+                "OPENRAG_API_KEY environment variable is required. "
+                "Create an API key in OpenRAG Settings > API Keys."
+            )
+
+        # MCP httpx client configuration (OPENRAG_MCP_*)
+        self.mcp_timeout = _parse_float("OPENRAG_MCP_TIMEOUT", 60.0)
+        self.mcp_max_connections = _parse_int("OPENRAG_MCP_MAX_CONNECTIONS", 100)
+        self.mcp_max_keepalive_connections = _parse_int(
+            "OPENRAG_MCP_MAX_KEEPALIVE_CONNECTIONS", 20
+        )
+        self.mcp_max_retries = _parse_int("OPENRAG_MCP_MAX_RETRIES", 3)
+        self.mcp_follow_redirects = _parse_bool("OPENRAG_MCP_FOLLOW_REDIRECTS", True)
+
+    @property
+    def headers(self) -> dict[str, str]:
+        """Get HTTP headers for API requests."""
+        return {
+            "X-API-Key": self.api_key,
+            "Content-Type": "application/json",
+        }
+
+
+_config: Config | None = None
+_openrag_client: OpenRAGClient | None = None
+
+
+def get_config() -> Config:
+    """Get singleton config instance."""
+    global _config
+    if _config is None:
+        _config = Config()
+    return _config
+
+
+def get_openrag_client() -> OpenRAGClient:
+    """Get singleton OpenRAGClient instance."""
+    global _openrag_client
+    if _openrag_client is None:
+        # OpenRAGClient reads OPENRAG_API_KEY and OPENRAG_URL from env
+        _openrag_client = OpenRAGClient()
+    return _openrag_client
+
+
+def get_client():
+    """Get an httpx async client configured for OpenRAG.
+
+    This is kept for backward compatibility with operations
+    not yet supported by the SDK (list_documents, ingest_url).
+    """
+    import httpx
+
+    config = get_config()
+    return httpx.AsyncClient(
+        base_url=config.openrag_url,
+        headers=config.headers,
+        timeout=config.mcp_timeout,
+        limits=httpx.Limits(
+            max_connections=config.mcp_max_connections,
+            max_keepalive_connections=config.mcp_max_keepalive_connections,
+        ),
+        transport=httpx.AsyncHTTPTransport(retries=config.mcp_max_retries),
+        follow_redirects=config.mcp_follow_redirects,
+    )