Klavis
Open-source MCP integration platform with 100+ prebuilt OAuth-backed connectors, intelligent routing, and sandboxes.
Delv Safety Grade: C
Score 58/100 · assessed 2026-04-18
Klavis is a community-built MCP integration platform claiming 100+ OAuth connectors and sandboxing capabilities. The project appears relatively new with limited maintainer track record - Klavis AI is not an established vendor. Docker-only distribution with no package manager support raises supply chain concerns, as users must build and run containers from source without version pinning or signed releases. The platform's scope is exceptionally broad, acting as a meta-layer that routes to numerous external services, which multiplies attack surface. While the open-source nature and documentation provide some transparency, the combination of wide permissions (OAuth to 100+ services, network access, potential filesystem access via Docker), unproven maintainer, and non-standard distribution creates meaningful risk. No known incidents, but the architecture inherently requires trusting both Klavis code and all connected services.
Green flags
- Open source on GitHub with visible code
- Sandboxing mentioned in description suggests security awareness
- OAuth-backed connectors better than API key sprawl
- No known security incidents to date
Red flags
- Docker-only install with no package manager or signed releases
- Claims 100+ OAuth connectors - extremely broad attack surface
- New/unproven maintainer with limited track record
- Meta-platform architecture multiplies trust requirements
- No versioned releases or dependency pinning visible
Permissions requested
Install
Self-hosted via Docker
Review
Install Klavis if you're connecting agents to multiple SaaS APIs and want OAuth handled properly. Skip it if you only need one or two integrations, or if Docker feels like too much overhead for your use case. It's a serious tool for serious multi-integration work.
Good at
- Bundles 100+ connectors into one MCP server, so you configure once instead of managing dozens of separate servers.
- OAuth flows are handled through a local web UI, no manual token copying or environment variable juggling.
- Intelligent routing means your agent can ask for actions in plain language without you hardcoding tool names.
- Sandboxed connectors isolate failures, so one broken integration doesn't take down the whole gateway.
- Self-hosted, so your API tokens stay on your infrastructure instead of passing through a third-party service.
Watch out
- Docker-only deployment adds overhead if you're not already running containers locally.
- Documentation for niche connectors is patchy; you'll need to read source code for some integrations.
- The routing logic is powerful but not immediately obvious, expect a learning curve before it clicks.
- Young project, so expect rough edges and occasional breaking changes as it matures.
- Overkill for simple use cases; if you only need one or two integrations, individual MCP servers are faster to set up.
Use cases
- multi-integration hub
- OAuth connector bundling
- agent tool routing
- enterprise MCP gateway
Getting started
Works with
Similar MCPs
- Lara Translate MCPProfessional translation with language detection, context-aware translations, translation memories, and glossaries.
- Make MCPTurns Make automation scenarios into callable tools for AI assistants, bridging 1,000+ apps through Make workflows.
- CCXT MCPCrypto exchange integration using CCXT, covering spot, futures, OHLCV, balances, and orders across 20+ exchanges.
- HubSpotHubSpot's MCP for CRM workflows: contacts, deals, tickets, properties. Distributed via NPM, see the developer docs for install steps.