Gemini CLI speaks MCP, so you can connect your ContextBolt library and search everything you have saved on X, Reddit, and LinkedIn from the terminal.
The settings.json entry
Open ~/.gemini/settings.json for a global setup, or .gemini/settings.json in a repo root to scope it to one project:
{
"mcpServers": {
"bookmarks": {
"httpUrl": "https://api.contextbolt.app/mcp/YOUR_TOKEN"
}
}
}
The field is httpUrl, not url. Gemini CLI is the odd one out here: Cursor, Claude Code, and Kiro all use url, Antigravity and Windsurf use serverUrl, VS Code uses url with an explicit type, and Gemini CLI wants httpUrl.
A config copied from any of those parses fine and gives you no tools. If nothing shows up after setup, this field is the reason far more often than anything else.
Confirm with /mcp
Start Gemini CLI and run /mcp. It lists every connected server along with the tools each one exposes.
This is more reliable than asking the model what tools it has, because a model can happily tell you it has bookmark access when the server never loaded. /mcp reads the actual connection state.
If /mcp lists it but queries come back empty
MCP reads from your synced library. With cloud sync switched off in the extension, everything looks correct: the server connects, /mcp lists it with all its tools, and every search returns nothing.
Turn sync on in the extension settings, give it a minute to upload, then retry. Check this before you start editing the config, because the symptoms point somewhere else entirely.
Working patterns
Ground the task first: ask it to check your bookmarks for relevant patterns before it scaffolds anything, so what it generates follows approaches you already chose.
Pick up dropped threads: get_recent_bookmarks answers “what did I save this month and never read” in one question.
See what you actually collect: list_clusters gives an honest picture of where your reading attention has gone.
Because your MCP URL carries a personal token, keep a per-project .gemini/settings.json out of any public repo. The ContextBolt bookmarks overview covers what the library does, and connect Claude to your bookmarks walks the same setup in a different agent.
How to connect ContextBolt to Gemini CLI
-
Get your MCP token from ContextBolt
Open the ContextBolt extension popup and click Settings. Under the MCP section, click Copy MCP Token. MCP is a Pro feature ($6/month) and cloud sync must be enabled in the same panel, otherwise the server returns nothing.
-
Add it to settings.json
Open ~/.gemini/settings.json for a global setup, or .gemini/settings.json in your repo root for a per-project one. Under the mcpServers object, add bookmarks with an httpUrl field pointing at your MCP URL.
-
Use httpUrl, not url
Gemini CLI uses httpUrl for a remote HTTP server. A url field is treated differently and your tools will not appear. This is the single most common failure and the reason a config copied from Cursor or Claude Code does not work here.
-
Run /mcp to confirm
Save the file, start Gemini CLI, and run /mcp. It lists every connected server and the tools each exposes, so you can confirm bookmarks is live before relying on it. Then ask 'search my bookmarks for anything about onboarding'.
Example prompts for ContextBolt + Gemini CLI
Once connected, try asking Gemini CLI:
Semantic search across X, Reddit, and LinkedIn at once, without leaving the terminal.
Puts material you already vetted into context so the generated code follows patterns you chose.
Calls get_recent_bookmarks with a window, good for picking up threads you flagged and never returned to.
Calls list_clusters, which gives you an honest picture of what you have actually been collecting.
What you can do with ContextBolt in Gemini CLI
Search saved tweets, Reddit posts, and LinkedIn content by meaning
Pull saved material into a task before code gets written
Filter by platform when you remember where you saved something
Browse your automatic topic clusters
Save new links to your library from the terminal
Confirm the connection any time with /mcp