Kollektiv MCP
About
Build and access a personal LLM knowledge base from your editor or client without any infrastructure setup.
Details
- Author
- alexander-zuev
- Categories
- Database, AI
Jump to
Setup
Install Kollektiv MCP in your MCP client (Claude Desktop, Cursor, Windsurf, and others).
Repository: https://github.com/alexander-zuev/kollektiv-mcp
Follow the installation instructions in the repository README, then restart your MCP client.
🚨 IMPORTANT: This experimental MCP server is now DEPRECATED and will be shut down soon.
Please do not use this server for new projects.
🧠 Your personal LLM knowledgebase(DEPRECATED)
[Original Description - No longer maintained]Kollektiv MCP enables you to build personal LLM knowledge base in seconds and use it from your favorite editor / client. No more infrastructure setup, chunking, syncing - just upload your data and start chatting. Supports all major MCP clients out of the box - Cursor, Windsurf, Claude Desktop, etc.
This experimental MCP server isDEPRECATEDand will beshut down soon. The service endpoints may stop working at any time without notice.
Do not use this for new projects or production use.
💿 Connection (DEPRECATED - MAY NOT WORK)
The simplest way to connect to Kollektiv MCP is to copy & paste the following configuration into your editor'smcp.jsonfile. All clients (Cursor, Windsurf, Claude Desktop, VSCode, PyCharm) support thisjsonformat
{ "mcpServers": { "kollektiv": { "command": "npx", "args": [ "-y", "mcp-remote", "https://mcp.thekollektiv.ai/mcp" ] } } }
- name:
- kollektiv- you can you give the server any descriptive name
- npx- ensure you have node.js install before running this command
- -y- this enables your shell to installmcp-remotewhich is currently required to connect to remote servers
- mcp-remote- this enables your client to connect to a remote MCP server (in this case Kollektiv)
- https://mcp.thekollektiv.ai/mcp- is the endpoint you are connecting to
Check out a short demo below or read client-specific instructions on how to connect.
Open Cursor and go toCursor Settings > MCP > Add new global MCP Server. Paste the configuration above and save (ctrl/cmd+s).
If configuration is successful and you haven't authenticated before, a browser window should open guiding you to the login page.
💡After saving thejsonit might take a while for Cursor to connect to the MCP. You might need to restart Cursor or give it a bit of time. If you see 'Client is closed' or other errors, taking thesetroubleshootingsteps might help.
If the connection is successful, you should see Kollektiv MCP go green in the settings page:
Open Windsurf and go toSettings -> Windsurf Settings > MCP Servers > View raw config. Paste the configuration above and save (ctrl/cmd+s).
If configuration is successful and you haven't authenticated before, a browser window should open guiding you to the login page.
💡Windsurf, in contrast to other clients, in my experience requires a restart of the app to properly connect. If the server doesn't go 'green' after a while, try going over thetroubleshootingsteps below.
If connection is successful you should see Kollektiv MCP go green in the settings page:
Open Claude Desktop and go toSettings -> Developer > Edit config. Openjsonfile in any text / code editor, paste the configuration above and save (ctrl/cmd+s).
If configuration is successful and you haven't authenticated before, a browser window should open guiding you to the login page.
💡Claude for Desktop requires a restart of the app to properly connect. If the server doesn't go 'green' after a while, try going over thetroubleshootingsteps below.
If connection is successful you should see Kollektiv MCP go green in the settings page:
Open VS Code and go toSettings -> MCP: Add server > Command (stdio):
- command:
- npx -y mcp-remote https://mcp.thekollektiv.ai/mcp
- give your server a descriptive name such askollektiv
Your configurationsettings.jsonshould look similar to this:
{ "chat.mcp.discovery.enabled": true, "chat.mcp.enabled": true, "mcp": { "servers": { "kollektiv": { "type": "stdio", "command": "npx", "args": [ "-y", "mcp-remote", "https://mcp.thekollektiv.ai/mcp" ] } } } }
- ClickStartto connect to the MCP Server
- if you are not authenticated - you will be taken to the authentication page
💡VS Code requires you to manuallystartyour server, addchat.mcp.enabledand switch to Agent mode to use MCP. If you do not see MCP tools in Agent mode, try going over thetroubleshootingsteps below.
If connection is successful you should see the tools exposed by Kollektiv MCP.
Open Cline, click onMCP Servers > Edit Configurationand add the following configuration to yourcline_mcp_settings.json:
{ "mcpServers": { "kollektiv": { "timeout": 60, "command": "npx", "args": [ "-y", "mcp-remote", "https://mcp.thekollektiv.ai/mcp" ], "transportType": "stdio", "disabled": false } } }
Note: direct connections to remote servers that support Authorization are not supported yet by Cline.
If connection is successful, you will be taken through to the authentication flow. After logging in, you should see Kollektiv MCP enabled in Cline.
Most MCP clients follow the same.jsonformat and should work with similar configuration steps as previously mentioned clients:
- Copy & paste configuration into your client'sjsonconfiguration
- Restart the app
- Authenticate if not already
- Kollektiv MCP should go green & be available in chat / agent mode
Success of your connection depends on many factors including but not limited to:
- how strong developers of a particular client wanted to support MCP connections
- whether the client supports the latestMCP specwith Oauth support
If you are experiencing issues, going through these simpletroubleshooting stepsmight help.
I've validated connection works to the following MCP clients:
- Cursor ✅
- Windsurf ✅
- Claude Desktop ✅
- VS Code ✅
- Cline ✅
Other MCP clientsshouldbe supported in theory, but in practice things might be a little different. If you have a client you really want to connect to - let me know!
- /query_documents— Submit a question to the documents you’ve uploaded to Kollektiv and receive an answer based on the sources from your documents.
- /list_documents— Return a list of your synced documents together with basic metadata.
- Pro tip:Include the phrase“use Kollektiv MCP”so the client knows to call these tools.
- Always add "use Kollektiv MCP"— This tells the client which MCP server to use.
- Wait for document to be Available— After upload, it takes 1–2 minutes before the document can be queried.
- Rephrase queries when needed— If the client generates a poor query, edit or rewrite it yourself.
This MCP server uses Cloudflare Agents SDK as well as other libraries to provide the most modern way for users to connect to and use MCP servers. MCP clients on the other hand have yet to implement support for the 2 critical pieces:
- remote MCP servers
- MCP server authorization
In case you experience connection issues, please go through the following troubleshooting steps which should help you connect to the MCP server.
If you require additional support please open a GitHub issue or reach out atsupport@thekollektiv.ai
If you are gettingInvalid Authorization Requesterror as below or can not connect for another reason, try going through the steps below which should fix the issue.
-
Ensure you're connecting to the correct endpoint:
- Usehttps://mcp.thekollektiv.ai/mcpas the MCP endpoint.
- What this does:
- Removes cache ofmcp-remotelibrary that is used to connect to the remote server from a client that doesn't support remote connections
- Run the following command in your terminal
# MacOS rm -rf ~/.mcp-auth # Windows Remove-Item -Recurse -Force "$env:USERPROFILE\.mcp-auth"
- Open your browser settings and delete browsing data for the last several hours
⚠️ Note: this will sign you out from all active sessions, including Kollektiv. Only do this if you're stuck in a broken login flow.
- Restart your MCP client and try to reconnect to the MCP server:
- What this does:
- MCP clients (Cursor, Windsurf, etc.) often cache connection / configuration settings from previous runs which might interfere with authentication.
- Restart your editor / client
- Try reconnecting to the MCP server
For debugging purposes, you can use the MCP Inspector to connect to Kollektiv MCP server.
Select either SSE or Streamable HTTP transport
- SSE: connect to the server athttps://mcp.thekollektiv.ai/sse
- Streamable HTTP: connect to the server athttps://mcp.thekollektiv.ai/mcp
If you're just here for Kollektiv - skip this. This section is for devs and builders curious about how it works.
Kollektiv MCP is part of a modular system enabling users to set up RAG over their data in seconds — without the need to manage infrastructure, pipelines, or model configs.
It consists of three independently deployed services:
-
MCP Server (Cloudflare Worker)
https://mcp.thekollektiv.ai
Acts as a secure gateway for clients to interact with indexed data via the Model Context Protocol. Supports OAuth.
Frontend (React + Vite Worker)
https://thekollektiv.ai
A clean, minimal user interface for uploading and managing their content.
Backend (FastAPI)
https://api.thekollektiv.ai
Handles source ingestion, validation, and orchestration of a RAG pipeline.
Kollektiv MCP implements several security measures:
All traffic is servedexclusively over HTTPS through Cloudflare’s edge, and every sensitive POST request carries a one-time CSRF/transaction token.
The backend runs inside theCloudflare Workers sandbox(no local file-system, no long-running processes), drastically reducing the attack surface.
For detailed disclosure guidelines seeSECURITY.md.
Released under the Apache License 2.0 — commercial support or alternative licensing:azuev@outlook.com
Hydrolix time-series datalake integration providing schema exploration and query capabilities to LLM-based workflows.
614 million contacts, live SMTP-verified. Invalid emails cost you $0. Connect Argorant with any Agent through CLI, MCP or API and build verified lists in Claude, GPT or Hermes.
Chinese writing toolbox for Claude Code: text analysis, outline generation, GB/T 7714 references,character frequency
A lightweight MCP server for connecting to Apache Doris and other MySQL-compatible databases, providing tools and prompts for LLM applications.
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.



