Choosing an MCP gateway: how Kravn compares
If you're evaluating how to bring the Model Context Protocol to your organization, you're weighing a few real options: a SaaS product that hosts everything for you, a general-purpose self-hosted project, or a gateway like Kravn that is self-hosted and built around enterprise identity and governance. This page lays out what to look for and where Kravn fits — honestly, including where it isn't the right tool.
What an MCP gateway actually needs to do
Before comparing products, compare on the dimensions that matter once MCP is more than a demo:
| Dimension | The question to ask |
|---|---|
| Single governed surface | Can one endpoint front many upstream MCP servers, so clients point at the gateway — not a dozen scattered URLs? |
| Identity & access | Does it plug into your IdP (SAML, OIDC), provision users (SCIM), and enforce per-team access to servers and individual tools? |
| Data boundary | Where do prompts, context and results go? Can you guarantee nothing leaves your network? |
| Deployment & ops | How long from zero to running? Is day-2 config done in-app, or does every change mean a redeploy? |
| Governance & audit | Can you redact secrets/PII, guard against prompt injection, and keep a tamper-evident audit trail? |
| Extensibility | Can you add integrations and lifecycle hooks without operating a separate server per source? |
| Licensing | Can you read the source, self-host without a per-seat bill, and avoid lock-in? |
Kravn was built to answer all seven with a yes — that's the whole design goal.
Kravn vs. the common approaches
| SaaS AI tooling | Generic self-hosted | Kravn | |
|---|---|---|---|
| Runs on your infrastructure | ❌ vendor cloud | ✅ | ✅ Docker / Helm, your network |
| Data leaves your network | ⚠️ usually yes | varies | ✅ no egress by design |
| Enterprise identity (SSO / SCIM / RBAC) | varies | ⚠️ often thin | ✅ SAML, OIDC, SCIM 2.0, RBAC, teams |
| Per-team tool entitlements | rare | ⚠️ rare | ✅ per-team MCP + tool grants |
| Governance pipelines & audit | ⚠️ limited | ⚠️ rare | ✅ redact PII/secrets, prompt-injection guard, tamper-evident audit |
| Time to running | instant (hosted) | hours | ✅ one command, first-run wizard |
| Day-2 config without redeploy | n/a | ⚠️ config files | ✅ edited at runtime in Settings |
| Licensing | proprietary | varies | ✅ source-available (BSL 1.1 → Apache 2.0) |
The middle column is a generalization of self-hostable projects, not any single product — evaluate your specific shortlist on the same rows.
Kravn vs. IBM MCP Context Forge
IBM's MCP Context Forge is an open-source (Apache-2.0) MCP gateway and registry, built in Python. It's broad and mature, and Kravn is unapologetically inspired by it — Kravn is a leaner, opinionated take in a different stack:
- Stack: Kravn is a TypeScript monorepo (Fastify + Vue 3), shipped as a single image that boots with
docker compose up. No Python runtime to operate. - Focus: Kravn leads with corporate identity and governance — SSO/SCIM/RBAC, per-team entitlements, governance pipelines and a tamper-evident audit trail as first-class features, with a first-run setup wizard and runtime config so day-2 changes don't need a redeploy.
- Scope: Context Forge covers more surface area; Kravn deliberately trades breadth for a simpler, install-anywhere posture aimed at regulated and compliance-bound teams.
Both are self-hostable and source-visible. If you want the broadest Python-based reference implementation, Context Forge is a strong choice. If you want a compact, identity-first gateway that a bank can install without information leaving the building, that's Kravn.
Kravn vs. broader platforms (Obot, TrueFoundry)
Some tools that show up next to Kravn are broader platforms that happen to include gateway features:
- Obot is an AI-agent platform; the MCP gateway is one part of a larger agent-building product.
- TrueFoundry is an ML/LLM infrastructure platform (model deployment, an AI gateway) aimed at ML-platform teams.
Kravn is narrower on purpose: it is an MCP gateway, registry and proxy — not an agent builder or a model-serving platform. If your problem is "safely expose MCP tools to AI clients under corporate identity and governance, on our own infrastructure," that focus is the point.
When Kravn is the right fit
- You need MCP on your own infrastructure, with no data leaving the perimeter.
- Corporate identity and governance are non-negotiable: SSO, SCIM, RBAC, per-team entitlements, audit.
- You want to be running in one command and manage config in-app, not via redeploys.
- You value source-available software you can read, run and trust.
When it isn't
- You want a fully hosted, zero-ops SaaS and are fine with data leaving your network — a hosted product is simpler.
- You need an end-to-end agent-building or model-serving platform — that's a different category.
- You need the broadest possible feature surface today and don't mind operating a Python stack — evaluate IBM MCP Context Forge.
See for yourself
- Why Kravn — the positioning in full
- Get started — running in one command
- Security & compliance — the governance and licensing detail
- Frequently asked questions
- Source on GitHub