Teleport MCP Access – Zero Trust for AI Tools
Teleport MCP Access lets organizations protect Model Context Protocol servers with Teleport's identity-aware access controls and audit capabilities. It also lets MCP clients connect to Teleport-managed resources such as PostgreSQL databases through the tsh client, subject to each user's Teleport permissions.
Overview
Teleport MCP Access is Teleport's Model Context Protocol capability for securing MCP servers and exposing selected Teleport resources to AI clients. Instead of allowing an AI agent to reach internal MCP servers directly, Teleport places the connection behind its access-control layer so organizations can authenticate users, authorize tool access, and audit activity consistently with the rest of their infrastructure access program.
What the MCP server enables
Teleport documentation describes two closely related MCP workflows. First, administrators can enroll existing MCP servers behind the Teleport Application Service and grant access through Teleport RBAC. Teleport supports enrollment patterns for streamable HTTP, stdio, and SSE MCP servers, and it can authenticate to MCP servers with Teleport-signed JWTs that carry user identity, roles, and traits.
Second, Teleport can expose Teleport-managed PostgreSQL database resources to MCP clients. After a user signs in with tsh login, they can list available MCP servers with tsh mcp ls or generate client configuration with tsh mcp config. For database access, tsh mcp db config can configure Claude Desktop, Cursor, or a generic MCP client, and the resulting MCP tools include listing accessible databases and executing PostgreSQL queries through Teleport.
When to use it
Use Teleport MCP Access when MCP tools need enterprise access controls rather than direct network exposure. It is useful for security teams that want to govern which AI clients can reach internal MCP servers, which MCP tools a user can call, and which infrastructure resources a model can query. It is also useful when developers or analysts need natural-language access to Teleport-protected PostgreSQL data without sharing static database credentials with the AI client.
Connection and authentication
Client-side access is mediated by the tsh command-line client. A user signs in to the Teleport Proxy Service with tsh login, then the MCP client launches Teleport's local stdio bridge using tsh mcp connect <server-name> for enrolled MCP servers or tsh mcp db start <teleport-resource-uri> for Teleport database access. Teleport documentation states that MCP access requires a Teleport cluster and the tsh client, and that MCP commands require tsh version 18.1.0 or newer.
Key considerations
Permissions are central to safe operation. Teleport notes that users need permission for both the MCP server and the tools it exposes; missing tool permissions can result in an empty tool list. For database access, Teleport recommends creating a database user with only the permissions the model should have, such as read-only PostgreSQL access, because language models can execute queries through the MCP tool. Active tsh sessions are also required: if the session expires, MCP tool calls can fail until the user logs in again.
Supported Transports
stdio
Command: tsh
Args:
mcpconnect${TELEPORT_MCP_SERVER_NAME}
Frequently Asked Questions
- When should an AI agent use Teleport MCP Access?
- Use it when an AI agent needs to reach MCP servers or Teleport-managed PostgreSQL databases through Teleport's identity, RBAC, and audit controls instead of direct network access or static credentials.
- What does Teleport add to an AI agent's MCP capabilities?
- Teleport adds an access-control layer around MCP use. It lets users list and connect to MCP servers they are allowed to use, and it can expose Teleport-protected PostgreSQL databases through MCP tools such as database listing and PostgreSQL query execution.
- What can the AI agent access or manage through this provider?
- The agent can access MCP servers enrolled behind Teleport and, for Teleport database access, PostgreSQL database resources available to the signed-in Teleport user. Actual tools and resources depend on Teleport RBAC, enrolled resources, and the user's active session.
- How is authentication configured for Teleport MCP Access?
- Authentication is handled through the Teleport tsh client. The user signs in with tsh login to the organization's Teleport Proxy Service, then the MCP client launches tsh mcp connect or tsh mcp db start. Teleport can also authenticate to enrolled MCP servers with Teleport-signed JWTs when that enrollment model is configured.
- Which transport should be used for this MCP provider?
- Use the stdio transport from the MCP client to the local tsh command. Teleport can protect backend MCP servers that use streamable HTTP, SSE, or stdio, but the client configuration generated by tsh launches Teleport's stdio bridge with tsh mcp connect.