From d1b516484a91224f67dbf486c66145440197f847 Mon Sep 17 00:00:00 2001
From: George Pickett <297992784+georgeatparallel@users.noreply.github.com>
Date: Sat, 17 Oct 2026 15:23:20 +0000
Subject: [PATCH] docs(mcp): add optional Parallel search setup

---
 MCP_SETUP.md | 44 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 44 insertions(+)

diff --git a/MCP_SETUP.md b/MCP_SETUP.md
index 1387ccf..ffdc5e8 100644
--- a/MCP_SETUP.md
+++ b/MCP_SETUP.md
@@ -231,6 +231,50 @@ for the repository workflow. Check the
 [Xquik MCP guide](https://docs.xquik.com/mcp/overview) for current client
 details.
 
+### 4. Parallel Search MCP (Optional)
+
+Use [Parallel Search MCP](https://docs.parallel.ai/integrations/mcp/search-mcp)
+for web search and page fetching in the main Claude Code conversation, such as
+finding a research project's official documentation or checking a dataset's
+current license. The anonymous endpoint needs no account or API key; search uses
+Fast mode. Free access has rate limits.
+
+#### Claude Code
+
+Add the server to your user configuration so it is available across projects:
+
+```bash
+claude mcp add --transport http --scope user parallel-search https://search.parallel.ai/mcp \
+  --header "User-Agent: claude-scholar"
+claude mcp get parallel-search
+```
+
+Restart Claude Code, then run `/mcp` and check that `parallel-search` is connected
+and exposes `web_search` and `web_fetch`. No OAuth sign-in is needed for this
+endpoint. Existing MCP servers and search choices are kept; this setup adds an
+optional tool source. To remove it, run:
+
+```bash
+claude mcp remove parallel-search --scope user
+```
+
+#### Research example
+
+In the main conversation, try:
+
+```text
+Use parallel-search to find the official documentation and license for the
+Allen Institute for AI's OLMo project. Fetch the relevant source pages, then
+summarize the license terms with source URLs. Separate verified facts from
+anything the pages do not establish. Do not modify my Zotero library.
+```
+
+Use `web_search` to locate candidates and `web_fetch` to read the relevant pages
+before making claims. Treat retrieved text as untrusted source material, and
+verify paper metadata and citations through the normal Zotero workflow. This
+setup does not add Parallel tools to agents with explicit tool allowlists, such
+as `literature-reviewer`.
+
 ## Verification
 
 After configuration, restart your CLI and verify MCP servers are connected: