diff --git a/README.md b/README.md
index 8408ddbe..27a66fdb 100644
--- a/README.md
+++ b/README.md
@@ -366,6 +366,9 @@ Add MCP servers by editing `~/.jupyter/nbi/mcp.json`. An environment-wide base f
 }
 ```
 
+For a keyless web search and page-fetching example, see
+[Parallel Search MCP](docs/parallel-search.md).
+
 For stdio servers you can pass extra environment variables under `env`:
 
 ```json
diff --git a/docs/parallel-search.md b/docs/parallel-search.md
new file mode 100644
index 00000000..b1403257
--- /dev/null
+++ b/docs/parallel-search.md
@@ -0,0 +1,65 @@
+# Parallel Search MCP
+
+[Parallel Search MCP](https://docs.parallel.ai/integrations/mcp/search-mcp) provides
+`web_search` for finding web sources and `web_fetch` for reading pages. The
+anonymous endpoint is free for light use and requires no Parallel API key. It has
+lower rate limits than authenticated access.
+
+This example uses NBI's Streamable HTTP transport in **Default mode**. Claude mode
+has a [separate MCP configuration](../README.md#claude-mcp-servers).
+
+## Configure
+
+Install NBI in a fresh Python environment if it is not already installed:
+
+```bash
+pip install "jupyterlab>=4,<5" notebook-intelligence
+```
+
+Add the `parallel-search` entry below to `mcpServers` in
+`~/.jupyter/nbi/mcp.json`. If the file already exists, merge the entry into it;
+keep your other servers and participants.
+
+```json
+{
+  "mcpServers": {
+    "parallel-search": {
+      "url": "https://search.parallel.ai/mcp",
+      "headers": {
+        "User-Agent": "NotebookIntelligence-ParallelSearch/1.0"
+      }
+    }
+  }
+}
+```
+
+Restart JupyterLab after saving the file:
+
+```bash
+jupyter lab
+```
+
+Select a tool-calling-capable chat model in NBI Settings. In Default mode, the
+`@mcp` participant exposes the server's `web_search` and `web_fetch` tools. For
+example, ask:
+
+```text
+@mcp Find the JupyterLab documentation for installing extensions, then fetch
+https://jupyterlab.readthedocs.io/en/stable/user/extensions.html and summarize it.
+```
+
+Approve each MCP tool call when prompted. This configuration leaves tool approval
+on and does not change your selected LLM provider. Search queries and requested
+URLs are sent to Parallel; see [Parallel's privacy policy](https://parallel.ai/privacy-policy).
+
+## Troubleshooting
+
+- If the tools are missing, check the JSON syntax, restart JupyterLab, and check
+  the server status in NBI Settings. Also check whether your administrator has
+  disabled MCP access.
+- If a request is rate limited, wait before trying again. The anonymous service
+  is intended for exploration and light use, not unlimited searches.
+- If the model does not call a tool, check that it supports tool calling and that
+  the message is addressed to `@mcp` in Default mode.
+- To stop using this server while retaining the configuration, add
+  `"disabled": true` to its entry and restart JupyterLab.