QuilrAI
Back to Resources
Engineering

MCP Authentication: What the Spec Doesn't Cover

MCP 1.0 has no authentication standard. Here's how QuilrAI fills the gap with mutual TLS, per-tool permission scoping, and server identity verification.

6 min read
March 2026

MCP 1.0 defines a rich protocol for tool registration, capability negotiation, and message framing, but it is deliberately silent on authentication. The spec leaves server identity verification, client authentication, and per-tool permission scoping entirely to implementers. In practice, this means most MCP server deployments today have no authentication at all: any process that can connect to the server socket can register tools and receive tool results. For production enterprise deployments, this is a critical gap.

What Is the Authentication Gap in MCP 1.0?

The absence of authentication in the spec is not an oversight, it reflects the protocol's origin as a tool-calling framework for local development environments where the agent and the MCP server run on the same machine. But enterprise deployments increasingly run MCP servers as network services, and a network-accessible MCP server with no authentication is an open API that can be called by any process on the network. The tool registry is especially dangerous: an attacker who can register a malicious tool will have it offered to the model on every subsequent request.

How Does Mutual TLS Provide Server Identity?

Mutual TLS (mTLS) is the most reliable mechanism for establishing bidirectional identity between an MCP client and server. With mTLS, both the gateway and the MCP server present certificates from a shared certificate authority, the gateway verifies the server is who it claims to be, and the server verifies the gateway is an authorized client. This prevents man-in-the-middle attacks, rogue MCP server registration, and lateral movement attacks where a compromised service attempts to present itself as a trusted MCP tool provider.

What Is Per-Tool Permission Scoping?

Beyond server identity, enterprise deployments need per-tool permission scoping: the ability to restrict which agents can invoke which tools, with what arguments, under what conditions. This is analogous to OAuth scopes but applied at the tool call level. QuilrAI implements per-tool permissions as a policy layer in the MCP gateway, each tool registration includes a permission policy specifying which agent identities can invoke it, what argument patterns are allowed, and whether invocation requires an additional approval step for sensitive operations.

QuilrAI

How QuilrAI addresses this: The QuilrAI MCP Gateway implements mTLS for all server connections, maintains a certificate registry for authorized MCP servers, and enforces per-tool permission policies defined in the Guardian setup wizard, including approval workflows for tools flagged as sensitive.

Related Articles

Dig deeper

Secure your AI stack today

See how QuilrAI's Guardian Agent and LLM Gateway protect your AI deployment from the threats covered in this article.

Get a Demo