docs: add a Cursor workflow for web research and long-term memory

MCP/README.md

@@ -163,6 +163,13 @@ Add to your MCP JSON settings:
 }
 ```
 
+### Web research with long-term memory
+
+Pair SimpleMem with an optional web search server in Cursor to research current
+information, save selected findings with source URLs and retrieval dates, and
+recall them in later conversations. See the
+[Parallel Search MCP workflow](../docs/research-memory.md) for configuration and
+example prompts.
 
 ## How It Works
 

docs/research-memory.md

@@ -0,0 +1,88 @@
+# Web research with long-term memory
+
+Use SimpleMem and [Parallel Search MCP](https://docs.parallel.ai/integrations/mcp/search-mcp)
+together in Cursor to research current information and recall selected findings in
+later conversations. Cursor coordinates two independent MCP servers: Parallel
+searches the web and fetches pages, while SimpleMem stores and retrieves memories.
+Adding the search server does not make SimpleMem's memory queries browse the web
+automatically.
+
+## Set up Cursor
+
+1. Complete the [SimpleMem MCP setup](../MCP/README.md) and obtain your SimpleMem
+   token. You can use the hosted service or your existing self-hosted endpoint.
+2. Open `~/.cursor/mcp.json`, Cursor's global MCP configuration. Merge the
+   `parallel-search` entry below into your existing `mcpServers` object. Keep your
+   existing SimpleMem URL, token, other servers, and enabled/disabled choices.
+   If you are starting from scratch, the full example is:
+
+   ```json
+   {
+     "mcpServers": {
+       "simplemem": {
+         "url": "https://mcp.simplemem.cloud/mcp",
+         "headers": {
+           "Authorization": "Bearer YOUR_TOKEN"
+         }
+       },
+       "parallel-search": {
+         "url": "https://search.parallel.ai/mcp"
+       }
+     }
+   }
+   ```
+
+3. Replace `YOUR_TOKEN` with your SimpleMem token. Keep the file private. Parallel's
+   anonymous endpoint needs no account or API key; its free tier is intended for
+   exploration and light use, with search running in Fast mode. SimpleMem still
+   needs its own authentication and backend configuration.
+4. In Cursor's MCP settings, confirm both servers connect and their tools appear.
+   Parallel exposes `web_search` and `web_fetch`; SimpleMem exposes `memory_add`,
+   `memory_retrieve`, and `memory_query`, among other tools. See the
+   [Cursor MCP documentation](https://cursor.com/docs/context/mcp) for client setup
+   and tool approval controls.
+
+## Research, then save selected findings
+
+Start an Agent conversation with both servers enabled. For example:
+
+> I'm documenting Python virtual environments for our team. First use SimpleMem's
+> memory_retrieve to check for earlier team decisions about Python environments.
+> Then use Parallel's web_search to find the current official Python venv
+> documentation. Use web_fetch if the search excerpts are insufficient to explain
+> how to create and activate an environment on Linux and Windows. Show the source
+> URLs, retrieval date, and any differences from our saved decisions. Do not save
+> anything yet.
+
+Review the sources and choose what should persist. Then ask:
+
+> Save only the findings I selected using SimpleMem's memory_add, with speaker
+> "Research notes". Include the topic, a short self-contained finding, its source
+> title and full URL, and today's absolute retrieval date in the content. Treat
+> these as external research findings, not as my preferences or team decisions.
+
+Keep notes focused so later queries can find them. Put provenance in the `content`
+alongside each finding; `memory_add` has no separate source-URL field. SimpleMem
+compresses input into facts, so it is not a verbatim source archive. After saving,
+ask `memory_retrieve` for the topic and check that the finding, URL, and retrieval
+date survived. If provenance is missing, correct the note and verify it again;
+keep the original source separately when exact wording matters.
+
+## Recall and refresh
+
+In a later conversation using the same SimpleMem account, try:
+
+> Use SimpleMem's memory_retrieve to recall our Python virtual-environment
+> research. Show the stored source URLs and retrieval dates. Use Parallel to check
+> the current documentation before making a recommendation. Explain any changes
+> and ask me before saving updated findings.
+
+`memory_retrieve` returns stored entries for inspection; `memory_query` generates
+an answer from memory. Neither establishes that a saved fact is still current.
+Use web research to recheck time-sensitive findings. If no memory is returned,
+broaden the topic query and confirm you are using the same SimpleMem account.
+
+If a server fails to connect, check its own configuration: the bearer token above
+belongs to SimpleMem only. If Parallel returns a free-tier rate limit, follow the
+retry guidance in the response or try again later; do not treat the failed request
+as evidence for a finding.