kollektiv-mcp-gateway
Facilitates lightning-fast Retrieval-Augmented Generation (RAG) functionality atop user-supplied datasets. Users can ingest documents and interact with the system via diverse MCP client applications. Security is enforced through OAuth 2.0 authentication and all data exchange occurs over secure HTTPS channels, ensuring a private interaction plane.
Author

alexander-zuev
Quick Info
Actions
Tags
❌ OBSOLETE - Kollektiv MCP Endpoint
‼️ CRITICAL UPDATE: This experimental MCP infrastructure is now RETIRED and service termination is imminent.
Refer to kollektiv.sh for current information.
Cease usage for any novel or ongoing development endeavors immediately.
~~🗂️ Establish Your Bespoke LLM Repository Instantly~~ (DECOMMISSIONED)
[Original Purpose - No longer maintained] Kollektiv MCP facilitates the rapid construction of a personalized large language model knowledge repository. Utilize it directly from your preferred editing environments or client applications. It eradicates infrastructure provisioning, data chunking, and synchronization overhead—simply upload your corpus and initiate dialogue. Compatible natively with mainstream MCP clients, including Cursor, Windsurf, and Claude Desktop.
🚨 Service Retirement Notice
This experimental MCP server instance is DECOMMISSIONED and will be deactivated shortly. Service endpoints might cease operation without prior notification.
Avoid deployment for any new initiatives or mission-critical environments.
💿 Access Configuration (DEPRECATED - FUNCTIONALITY UNRELIABLE)
The most straightforward connection method involves embedding the subsequent configuration block within your client's mcp.json manifest. This structure is recognized by all supported clients (Cursor, Windsurf, Claude Desktop, VSCode, PyCharm).
{ "mcpServers": { "kollektiv": { "command": "npx", "args": [ "-y", "mcp-remote", "https://mcp.thekollektiv.ai/mcp" ] } } }
- name:
- Designated as
kollektiv; this identifier is user-configurable.
- Designated as
- command:
- Specifies
npx; requires a functional Node.js environment.
- Specifies
- args:
-y: Permits automatic installation of the requisitemcp-remoteutility.mcp-remote: Essential utility for bridging to remote MCP endpoints.https://mcp.thekollektiv.ai/mcp: The target URI for the gateway.
Review the brief demonstration below or consult client-specific guides for detailed connection protocols.
Client Integration Guides (Cursor, Windsurf, Claude Desktop, VS Code, Cline)
Connection procedures vary slightly per IDE/Client. Generally, the process involves:
1. Locating the MCP configuration section (often mcp.json or equivalent settings panel).
2. Injecting the above JSON block.
3. Saving changes (often requiring an application restart, especially for Windsurf/Claude Desktop).
4. Completing the OAuth 2.0 authentication challenge when prompted by a browser redirect.
For VS Code, note the requirement to explicitly enable the feature via "chat.mcp.enabled": true in settings.json and switch to the 'Agent' mode.
Other Environments (PyCharm, Claude Code)
Compatibility generally extends to other clients respecting the standard JSON format. Successful integration hinges on:
- Client adherence to the MCP connectivity specification.
- Support for remote server bridging and the necessary authorization mechanisms (OAuth).
If linkage fails, consult the diagnostics section below.
Validated Client Ecosystem
Confirmed operational integration with:
- Cursor ✅
- Windsurf ✅
- Claude Desktop ✅
- VS Code ✅
- Cline ✅
Other clients are theoretically compatible, pending practical implementation variances.
🎮 Operational Commands
Exposed Capabilities
/query_documents— Formulate an inquiry against your uploaded document index to yield a source-grounded response./list_documents— Retrieve metadata pertaining to your indexed and synchronized files.- Usage Cue: Include the directive “use Kollektiv MCP” to signal the client to invoke these specific tools.
Interaction Best Practices
- Explicit Invocation: Always preface tool usage with “use Kollektiv MCP”.
- Indexing Latency: Allow 1-2 minutes post-upload before attempting to query newly added data.
- Query Refinement: If the client formulates an ineffective prompt, manually edit or rephrase the request.
❓ Diagnostics & Assistance
This server leverages the Cloudflare Agents SDK. Connection issues frequently stem from incomplete implementation of two key features within client applications:
- Remote endpoint handling.
- Server-side authorization protocols.
If connectivity problems persist, follow the prescribed debugging steps.
Support Channels
For deeper assistance, submit a formal issue on GitHub or contact support@thekollektiv.ai
Connection Remediation Steps
If you encounter authorization failures or general connection stalls (e.g., viewing the displayed Authorization Error):
-
Endpoint Verification: Confirm the precise URI used is
https://mcp.thekollektiv.ai/mcp. -
mcp-remoteArtifact Clearance:- Purpose: Erases cached authentication tokens for clients using the remote bridging utility.
- Execution (Terminal): bash
MacOS/Linux
rm -rf ~/.mcp-auth
Windows (PowerShell)
Remove-Item -Recurse -Force "$env:USERPROFILE.mcp-auth"
-
Browser Credential Scrub:
- Purpose: Removes locally stored authentication cookies essential for the OAuth handshake.
- Caution: This action logs you out of all sessions, including Kollektiv.
- Action: Clear browsing data/cookies for recent activity via browser settings.
-
Client Process Cycling:
- Purpose: Forces the MCP client (e.g., Cursor) to reread configuration and re-establish the connection state.
- Action: Terminate and restart your IDE/client application, then attempt linkage again.
Utilizing the MCP Inspector for Debugging
For specialized tracing, connect the MCP Inspector utility:
bash npx @modelcontextprotocol/inspector
Select the appropriate transport protocol:
- SSE: Target URI
https://mcp.thekollektiv.ai/sse - Streamable HTTP: Target URI
https://mcp.thekollektiv.ai/mcp
🛠️ Architectural Overview (For Technical Reviewers)
Readers focused solely on client configuration may bypass this section.
Kollektiv MCP integrates into a distributed ecosystem designed for zero-setup RAG deployment.
-
MCP Server (Cloudflare Worker)
https://mcp.thekollektiv.ai
Serves as the secure interface, translating Model Context Protocol requests into secure actions against indexed content, incorporating OAuth enforcement. -
Frontend Portal (React + Vite Worker)
https://thekollektiv.ai
Provides a minimal UI for content management and ingestion. -
Ingestion Backend (FastAPI)
https://api.thekollektiv.ai
Orchestrates data processing, validation, and the execution of the underlying RAG pipelines.
🔐 Security Posture
Security is enforced via several layers:
- Authentication utilizes the standard OAuth 2.1 flow mediated by Supabase; credentials are never stored server-side, relying exclusively on short-lived, encrypted (
HttpOnly,Secure) session tokens. - Transport is strictly enforced over HTTPS via Cloudflare's edge network. Sensitive operations require a single-use CSRF token.
- The execution environment is the Cloudflare Workers sandbox, precluding filesystem access and long-lived processes, significantly minimizing potential attack vectors.
Consult SECURITY.md for comprehensive disclosure policies.
📜 Licensing Information
This project is distributed under the Apache License 2.0. Inquiries regarding commercial licensing or support should be directed to azuev@outlook.com
