Klavis AI vs Stacklok (2026)

Hosted MCP servers from Klavis cover more than a hundred applications, from Affinity and Airtable to Zendesk and Zoho Mail. Each one connects per end user through OAuth 2.0 or an API key, so an agent reads and writes in that user's own accounts. Tools are reached through each server's MCP URL, through Strata or through the Call Tool API for plain function calling. Strata is one MCP server that lets the agent discover categories and actions across every connected server before executing one.

Sourceklavis.ai/docs/mcp-server/overviewread 2026-09-22

Every tool call carries the identity of the person who made it. Administrators register MCP servers as connectors in the Connector Gateway, an identity aware endpoint for MCP tool calls, and grant them to directory groups. The gateway then brokers each connector's credentials on the caller's behalf. ToolHive runs each MCP server in an isolated container with its own permissions and secrets, and a Virtual MCP Server aggregates backends into one endpoint with tool filtering for a given agent.

Sourcedocs.stacklok.com/connector-gatewayread 2026-09-22

The product executes the tool calls an agent makes and returns the results. Sequencing, branching and handoff live in the customer's own framework, and Klavis publishes guides for LangChain and LangGraph, LlamaIndex, CrewAI, Agno, Mastra, Google ADK and the major model APIs. Strata's discover, browse and execute tools help the agent pick an action. Klavis has no workflow runtime of its own to sequence, branch, retry or route work, or to combine deterministic workflow nodes with autonomous agent steps.

Sourceklavis.ai/docs/ai-platform-integration/overviewread 2026-09-22

A Virtual MCP Server aggregates backend MCP servers behind one endpoint with static backend selection, tool filtering and renaming. The Connector Gateway exposes a toolset scoped to a group. Both route a call to the right server instead of running work. The platform runs no workflow or agent loop of its own, and there is no sequencing, branching or retrying across deterministic workflow steps and autonomous agent steps.

Sourcedocs.stacklok.com/connector-gatewayread 2026-09-22

Slack and Discord clients come alongside the hosted servers, so a person can put MCP tools to work from those chat channels. Klavis gives no setup guide for either client. It names no schedule, webhook, event subscription or inbound queue that starts work without a person asking.

Sourceklavis.ai/faqread 2026-09-22

A client or agent calling a tool or a model is what sets the platform to work. ToolHive runs the servers and the gateways broker the calls, both at the caller's request. No schedule, event, webhook or inbound queue starts an agent run, and nothing wakes an agent unless a person or client starts the work.

Sourcedocs.stacklok.com/toolhiveread 2026-09-22

MCP servers for Notion, Google Drive, Confluence and similar tools let an agent query those systems live, but Klavis builds no retrieval layer over the customer's own content. Strata finds tools, not documents, by browsing categories and actions instead of semantic search. Klavis names no ingestion, indexing, retrieval or RAG layer of its own, so there are no retrieved passages or sources to show.

Sourceklavis.ai/docs/concepts/strataread 2026-09-22

The vMCP optimizer searches tool descriptions to surface only the relevant tools for each request and cut token use. The registry catalogs MCP servers and skills. There is no index, graph or embeddings layer over the customer's own documents.

Sourcedocs.stacklok.com/toolhive/guides-vmcp/optimizerread 2026-09-22

Each end user's credentials and integration status are stored so tools act without asking for consent again. A sandbox holds application state for one training or evaluation run. Klavis names no memory layer that keeps an agent's context across runs. Its Mem0 server connects an agent to another vendor's memory product.

Sourceklavis.ai/docs/api-reference/user/get-user-integrationsread 2026-09-22

The product runs and governs MCP servers, skills and model traffic. None of its components (the ToolHive runtime, Registry Server, vMCP, Connector Gateway, AI Gateway and Enterprise Manager) persists an agent's state or a user's context across runs, so agents get no memory from it.

Sourcedocs.stacklok.com/toolhiveread 2026-09-22

Each end user authorizes an integration through OAuth consent before its tools act, and a server connection can be set read only through the API. Klavis Guardrails sits as a proxy between MCP clients and servers. It detects tool poisoning and prompt injection, enforces least privilege access and checks tool invocations against allowlists. Klavis names no approval step a person acts on before an action commits, whether at a node, workflow or policy layer.

Sourceklavis.ai/docs/enterprise-security/klavis-securityread 2026-09-22

Administrators limit what an agent may do instead of putting a person in the loop. They decide which MCP servers are registered as connectors and which directory groups may use them. Enterprise Manager policies can block servers that are not in the approved registry, and authorization policies limit each role's tools. The AI Gateway screens prompts for injection and scans requests and responses for card and personal data. The customer controls all of these limits, but no person reviews, approves or validates an agent action before it commits.

Sourcedocs.stacklok.com/connector-gatewayread 2026-09-22

The API enforces HTTPS with API key authentication. Each end user's credentials are scoped per integration and returned only to the key owner that controls the instance, and customers can run OAuth through their own white label OAuth apps. Connections can be set read only, and Klavis Guardrails enforces least privilege tool access. No attestation or certification is published. On security, Klavis points only to its built in OAuth, and it publishes terms, privacy and cookie policies but no trust center. Klavis names no SSO, roles or audit trail.

Sourceklavis.ai/docs/api-reference/overviewread 2026-09-22

Identity comes from Okta or Microsoft Entra ID, with SCIM provisioning into a directory of users, groups and subgroups. Virtual API keys are bound to a user's identity. Enterprise authorization uses fleet wide MCP roles in RBAC custom resources, with delegation at the namespace level, and Cedar policies apply to MCP servers.

Connector credentials are stored encrypted at rest in the directory. The distribution itself carries Sigstore signatures, SLSA provenance and SBOMs the customer can verify. Stacklok holds no third party attestation of its own operations and says none applies to a self hosted product that holds no customer data.

Sourcestacklok.com/security-centerread 2026-09-22

The API returns server instances, tools, users and authentication status, and users check authentication in the dashboard. Klavis names no log, trace, execution history or audit event of the tool calls an agent makes through it, and no export to a SIEM. Guardrails' real time threat detection acts on calls as they happen, and Klavis does not say whether it keeps a record the customer can inspect.

Sourceklavis.ai/docs/api-reference/overviewread 2026-09-22

The AI Gateway records structured audit events for model requests, detections and key actions and forwards them to the customer's SIEM. The Connector Gateway's tool usage view shows traffic by time range, connector and tool. An Enterprise Manager telemetry policy enforces OpenTelemetry settings across Stacklok clients, so tool calls are traced into the customer's own stack. A customer can inspect what an agent sent and received, keep audit events apart from traces, and export both to its own tools.

Sourcedocs.stacklok.com/ai-gateway/forward-audit-logsread 2026-09-22

Hosted servers run on Klavis's infrastructure with a stated 99.9% uptime SLA. There are two production servers, api.klavis.ai for the US and api.eu.klavis.ai for the EU. For on premises or custom deployments, every MCP server ships as a Docker image to run in the customer's own infrastructure, and Strata is open source. Customers can pick the US or EU region, or run the servers in their own environment.

Sourceklavis.ai/docs/concepts/mcpread 2026-09-22

Everything runs inside the customer's own infrastructure. Stacklok Enterprise deploys as a single umbrella Helm chart into the customer's Kubernetes cluster. Distributed deployments can span several clusters, and an air gapped install mirrors the chart from a private registry. Stacklok neither hosts nor processes customer data.

Sourcedocs.stacklok.com/platformread 2026-09-22

The catalog holds MCP servers and tools. They give an agent actions to take and do no work on their own when selected. The framework guides build example agents as tutorial code. Klavis names no ready made workflow, template or role specific agent of its own for a customer to adopt.

Sourceklavis.ai/docs/mcp-server/overviewread 2026-09-22

Agent skills ship with ToolHive. They are reusable bundles of tools, permissions and secrets that a customer browses in a registry and installs into its AI clients. A skill can be built from a local folder as an OCI artifact, then published, listed or deleted through the Registry Server's admin API. The CLI and desktop UI also install, distribute and manage skills.

Sourcedocs.stacklok.com/toolhive/guides-registryread 2026-09-22

Tools are served to whatever model the customer runs. The List Tools API returns a server's tools in several model formats, and integration guides cover OpenAI, Anthropic's Claude, Gemini, Mistral, Cohere, Together AI and Fireworks AI, each using the customer's own provider key. Klavis names no model of its own, so the model, the provider and the key are the customer's throughout.

Sourceklavis.ai/docs/ai-platform-integration/overviewread 2026-09-22

Customers and admins decide which model serves a request. The AI Gateway is a self hosted control point between AI clients and model providers. The customer connects OpenAI, Anthropic, AWS Bedrock, Azure OpenAI, Google Vertex AI and OpenAI compatible providers. It then maps model names onto them with a default route, weighted load balancing, priority failover, retries and timeouts, declared as AIGateway and AIPolicy resources.

Sourcedocs.stacklok.com/ai-gatewayread 2026-09-22

Outside callers reach Klavis through a REST API at api.klavis.ai with a published OpenAPI specification. Python and TypeScript SDKs cover servers, Strata, users, OAuth, white labeling and sandboxes. Klavis serves MCP as an addressed service, with each hosted server and each Strata instance at its own URL. The servers and Strata are open source for self hosting.

Sourceklavis.ai/docs/installationread 2026-09-22

Builders get ToolHive, an Apache 2.0 open source platform with a CLI, desktop UI and Kubernetes operator. Its Registry Server implements the MCP registry API for servers, skills and plugins. Stacklok Enterprise publishes a REST API reference for the Enterprise Manager, Connector Gateway and AI Gateway. It also publishes custom resource definitions for authorization and a management API for budgets.

Sourcedocs.stacklok.com/toolhiveread 2026-09-22

Isolated sandboxes of real applications let the customer test and train its agent. A Klavis Sandbox is loaded with a deterministic world state, and the agent works against it through MCP. The full state can be exported to compare against ground truth, and the sandbox resets for the next run. The Universe API acquires benchmark task environments, and its verify endpoint runs the task's check and returns whether the task passed, with the failure reason. Sandboxes can also be configured for the MCP-Atlas, Toolathlon and MCP-Mark benchmarks.

Sourceklavis.ai/docs/api-reference/universe/verify-taskread 2026-09-22

Before an MCP server goes to production, ToolHive's CLI can test and validate it. The registry applies review criteria before a server is listed. There are no scored test cases, pass rates or judge verdicts that give a readable result to compare one change against another.

Sourcedocs.stacklok.com/toolhive/guides-cli/test-mcp-serversread 2026-09-22

The Local Sandbox API starts a virtual machine hosting interconnected MCP servers, among them Playwright for a browser plus a terminal and Desktop Commander for the machine. It serves filesystem, terminal, browser and document workflows during agent training and evaluation. Klavis restricts it to use outside production and says nothing more about driving a browser through the hosted Playwright server. The hosted MCP servers and Strata work through application APIs, not interfaces.

Sourceklavis.ai/docs/api-reference/local-sandbox/acquireread 2026-09-22

The product governs and runs MCP servers and model traffic, and drives no interface itself. It operates no browser, desktop or remote computer session for an agent, and no headless fetch or third party browsing engine is wired in.

Sourcedocs.stacklok.com/toolhiveread 2026-09-22