Cursor was the first coding tool to ship first-party MCP client support. When you connect a knowledge base via MCP, Cursor can query your notes, research, and documentation directly from the chat panel — no copy-paste, no context switching, no losing your train of thought to find that one decision you wrote down three weeks ago.
If you have used Glassy's MCP server with Claude Desktop or any other MCP client, the setup for Cursor is nearly identical. The difference is the workflow: Claude Desktop is a conversation. Cursor is your editor.
Why this matters for developers
Developers live in context. You read a design doc, switch to the editor, forget a detail, switch back, search for it, switch back again. The context-switching tax is real — studies show it takes 15-20 minutes to fully refocus after an interruption. MCP eliminates the switch.
With Glassy connected to Cursor, you can:
- Query design decisions — "Why did we choose SQLite over Postgres?" Cursor searches your Glassy notes and returns the answer with a link to the original decision document.
- Pull API documentation — Capture API docs with the browser extension, then ask Cursor to reference them while writing integration code.
- Find research notes — "What did I save about WebGPU compute shaders?" Your captured articles and papers are one query away.
- Create notes from chat — Cursor can write decisions back to Glassy via the
create_noteMCP tool, building your knowledge base as you code.
Step-by-step setup
1. Generate an MCP key in Glassy
Open Glassy at glassy.fyi,
navigate to Settings → Integrations → MCP Server, and click
Generate MCP Key. Copy the key — it starts with
gky_mcp_.
2. Add Glassy to Cursor's MCP config
Open Cursor settings (gear icon → Settings), find the MCP section, and add a new server. Or edit
your ~/.cursor/mcp.json directly:
{
"mcpServers": {
"glassy": {
"url": "https://app.glassy.fyi/mcp",
"headers": {
"Authorization": "Bearer gky_mcp_your_key_here"
}
}
}
}3. Start coding
Restart Cursor. Open the chat panel. You should see Glassy listed as an available MCP server.
Try asking: "Search my knowledge base for notes about authentication patterns." Cursor will
call the search tool
and return results inline.
The developer workflow
Here is how a typical development session looks with Glassy + Cursor:
- Morning research — You capture three articles about a new library using the Glassy Companion extension. They are indexed automatically with hybrid search.
- Start coding — Open Cursor, ask the chat: "What do I know about the new library API based on my research?" Cursor queries Glassy via MCP and returns a summary with citations to your captured articles.
- Make a decision — You decide on an approach. Ask Cursor to create a note in Glassy: "Create a note titled 'Library X integration decision' with the key points from this conversation." The note is saved to your knowledge base via the
create_notetool. - Afternoon — A colleague asks why you chose that approach. You query Glassy from Cursor: "Find my note about the Library X integration decision." The note comes back instantly with the full reasoning.
What Cursor sees
Glassy's MCP server exposes fifteen tools to Cursor. The most useful for developers:
- search — Hybrid BM25 + vector search across your entire corpus. Returns ranked results with snippets and source URLs.
- read — Fetch the full content of a specific note or bookmark by ID. Returns Markdown.
- create_note — Write a new note from the Cursor chat. Great for saving decisions and code snippets.
- create_bookmark — Save a URL as a bookmark. Useful for linking to docs or repos.
- list_tags — Browse your tag catalog to discover what you have accumulated.
- get_stats — Workspace statistics — item count, index health, storage used.
For a full breakdown of all fifteen tools, see our post on how the Glassy MCP server works.
Rate limits
Free accounts get 120 MCP tool calls per hour. Pro accounts get 1,200. Most coding sessions use 5-10 queries, so the Free tier covers a full day of development. See our state of MCP in 2026 for context on how Glassy's rate limiting compares to other MCP servers.
Cursor vs Claude Desktop vs other MCP clients
All support MCP. The difference is the context:
- Claude Desktop — General-purpose conversation. Good for research, planning, and brainstorming. You ask questions and get answers.
- Cursor, Windsurf, OpenCode — Coding context. The AI sees your code, your files, your terminal. It can query your knowledge base and apply the results directly to the code it is writing.
- Hermes, Openclaw, Pi, and others — Any MCP-compatible assistant or agent can query your knowledge base the same way, from whichever interface you prefer.
Many developers use several. Claude Desktop for morning research and planning. Cursor for implementation. Glassy's MCP server works with all of them simultaneously — the stateless HTTP transport means there is no conflict.