Klavis MCP Platform
Author: Klavis-AI
Description: Klavis AI (YC X25) MCP integration platform for running and managing MCP servers at scale. Offers (1) Strata: intelligent connectors optimized for context window, (2) 100+ prebuilt MCP integrations with OAuth support, and (3) MCP Sandbox: scalable MCP environments for LLM training/RL. Quick start options: cloud-hosted via klavis.ai; self-host MCP servers via Docker images (e.g., ghcr.io/klavis-ai/github-mcp-server:latest); install the open-source Strata CLI locally via `pipx install strata-mcp` and add servers via `strata add --type stdio ...`; or provision server instances via Python/TypeScript SDKs or REST API (api.klavis.ai).
Stars: 5.7k
Forks: 539
License: Apache License 2.0
Category: Enterprise
Overview
Installation
docker pull ghcr.io/klavis-ai/github-mcp-server:latest
docker run -p 5000:5000 ghcr.io/klavis-ai/github-mcp-server:latestpipx install strata-mcp
strata add --type stdio playwright npx @playwright/mcp@lateststrata add
Add an MCP server/connector to a local Strata installation (e.g., add a stdio Playwright MCP server via npx).
klavis_client.mcp_server.create_strata_server
Python SDK method to create a Strata server instance for a user with a list of MCP servers (e.g., Gmail, Slack).
klavis.mcp_server.create_server_instance
Python SDK method to create an individual MCP server instance (non-Strata) for a given server name and user.
klavis.mcpServer.createStrataServer
TypeScript SDK method to create a Strata server instance for a user with specified MCP servers.
klavis.mcpServer.createServerInstance
TypeScript SDK method to create an individual MCP server instance for a given server name and user.
POST https://api.klavis.ai/v1/mcp-server/strata
REST API endpoint to create a Strata server for a user with a list of servers.
POST https://api.klavis.ai/v1/mcp-server/instance
REST API endpoint to create an individual MCP server instance for a user and server name.
FAQs
What are the security implications of the unresolved GitHub Issue #712, and should it block production deployment of Klavis?
Without access to Issue #712's details, the security implications cannot be assessed. Best practice is to review the issue thread directly, request a vendor statement on timeline and impact, and conduct a security audit of your specific deployment. For risk-averse production environments, consider sandboxing Klavis behind network controls or deferring deployment until the issue closes with documented resolution.
How does Klavis handle OAuth token refresh and expiry across multiple connected services in a multi-tenant application?
Klavis manages OAuth server-side and returns service-specific authorization URLs per user, but the content does not detail automatic token refresh mechanics, expiry notifications, or error handling when tokens become invalid. Teams should verify whether refresh tokens are stored server-side, how expiry events surface to applications, and whether manual reauthorization flows are required when long-lived tokens expire.
Can I mix self-hosted Docker MCP servers with cloud-hosted Strata endpoints in the same agent configuration?
Yes, you can run both deployment models simultaneously. Most MCP clients accept multiple server entries in their configuration file, so you can point one entry to your local Docker server endpoint and another to the cloud Strata URL. Each server appears as an independent tool source to the agent, and the client handles routing calls appropriately.
What are the current limitations and known gotchas when using Klavis Strata with Cursor or VS Code Copilot since there are no dedicated setup guides?
Both editors use proprietary MCP client implementations with undocumented SSE-transport quirks. You'll need to manually map Strata's URL format to each editor's config schema, test authentication headers pass correctly through their proxy layers, and verify tool-discovery pagination doesn't break when the editor's MCP adapter expects flat schemas. Community Discord threads suggest Cursor sometimes strips query parameters from SSE endpoints.
How does Klavis Strata's progressive tool discovery work step by step, and how does it reduce context window usage compared to a flat tool list?
Strata's progressive discovery works in four phases: agents query service categories (Gmail, Slack, GitHub), request action types within a service (read, write, search), fetch detailed schemas for selected actions, then execute. This staged approach loads only relevant tool definitions at each step, versus flat lists injecting complete schemas for fifty-plus services upfront, consuming thousands of tokens before tool selection.
How does Klavis MCP Platform compare to Composio for teams that need both broad integration coverage and progressive discovery?
Klavis prioritizes progressive discovery through Strata's category-then-action-then-spec reveal, reducing prompt bloat for many simultaneous connections. Composio prioritizes catalog breadth with 500-plus integrations versus Klavis's 50-plus. For workflows hitting over fifty services or requiring compliance governance, Composio's wider catalog and published pricing aid long-term planning. For context budget and multi-step tool selection bottlenecks, Klavis delivers immediate relief.