MCP Security

Connect agents to your servers without sharing their credentials

Meandr MCP signs in to each MCP server on your agents' behalf. It keeps the secret encrypted where nobody can read it back, and checks who is on the other end of each connection.

Try for free Quick tour video

How it works

How MCP security works in Meandr

Meandr works as an MCP security gateway, from adding a server to the call that reaches it.

  1. 01 Authentication

    Meandr signs in to each MCP server the way that server expects

    Meandr handles MCP authentication to each server with one of nine methods, set when you add the server. Your agents never see the server's credential.

    • OAuth: Meandr renews the token before it expires; if it has already expired, the call is not sent and the agent gets a credentials stale error
    • AWS SigV4: Meandr signs each request over the full body it sends
    • mTLS: Meandr presents your client certificate in the TLS handshake
    Why SigV4 needs the body before signing How agents sign in to Meandr

  2. 02 Credentials

    Each server's credential is sealed, and nobody can read it back

    MCP credential management in Meandr: each credential is sealed with AES-256-GCM and decrypted in memory only to send a call to that server. Neither the dashboard nor the API returns the secret.

  3. 03 Connections

    Connections to your servers are encrypted, and each server has to prove who it is

    Meandr reaches each remote server over TLS and checks its certificate, so calls go only to secure MCP servers that prove who they are. For a server behind your own CA, turn on Custom trust roots and add the CA. With mTLS, the check runs both ways.

    • The certificate check is on by default, and only an explicit per-server flag turns it off
    • Your CA is added to the system roots, so public servers still verify
    • An stdio server connects through the Meandr CLI, which dials out to Meandr over TLS, so nothing on your machine accepts incoming connections
    • The CLI's tunnel token is shown once and removed from the environment before your server starts
    Guide: run an stdio server through the CLI

Questions about MCP security

Can I see a server's credential after I save it?
No. If a credential is wrong or has leaked, revoke it in the server's settings.
What is MCP authorization, and where does Meandr fit?
MCP authorization is how an MCP client gets permission to call a server, using OAuth 2.1 with PKCE. Claude Desktop signs in to Meandr this way, and Meandr signs in to each server with that server's own method. How agent keys work →
Is this the same as Meandr's own security?
No. What Meandr stores about your calls, for how long, and its SOC 2 status are on Security & trust. Security & trust →

Let Meandr hold the keys, so your agents don't

Add a server, pick how Meandr signs in to it, and point your agents at one endpoint.

Try for free

Questions? Email us