Snowflake MCP Server

Official, Snowflake-managed MCP server: Cortex Analyst, Cortex Search, and SQL execution as tools for AI agents, with no infrastructure to run yourself.

Description

The Snowflake-managed MCP server is the official Model Context Protocol server for the Snowflake data cloud platform, fully hosted and operated by Snowflake itself. Unlike most entries in this catalog, it is not a downloadable package that runs locally or in your own container: it is created directly inside a Snowflake account via SQL and is then reachable through an account-specific HTTPS endpoint. According to Snowflake, this lets AI agents securely retrieve data and functionality from a Snowflake account "without needing to deploy separate infrastructure." Per the official release notes, the feature reached general availability on November 4, 2025.

How the server is created

A Snowflake MCP server is a database object. It is created with the SQL statement CREATE MCP SERVER <name> FROM SPECIFICATION $$ ... $$, inside a schema; the specification is a YAML structure describing a list of tools. Each tool has a name, a type, and a description intended for the model. Per the documentation, a single server supports up to 50 tools across all types combined.

Tool types per the documentation

Five tool types are available: CORTEX_ANALYST_MESSAGE attaches a semantic view for structured, natural-language queries; CORTEX_SEARCH_SERVICE_QUERY exposes a Cortex Search service for searching unstructured text; SYSTEM_EXECUTE_SQL allows direct execution of model-generated SQL; CORTEX_AGENT_RUN invokes an existing Cortex Agent; and GENERIC wraps a user-defined function or stored procedure. This combination lets a single server expose structured metrics, unstructured documents, and free-form SQL together.

Access, authentication, and governance

The endpoint follows the pattern https://<account>.snowflakecomputing.com/api/v2/databases/<database>/schemas/<schema>/mcp-servers/<server_name>. For production access from MCP-capable clients, Snowflake defaults to its own OAuth 2.0 service; alternatively, the server can be bound to an external identity provider such as Okta or Microsoft Entra ID. For local development and initial testing, a Programmatic Access Token (PAT) used as a bearer token is the common approach. Governance runs through Snowflake's regular role-based access control (RBAC): creating a server requires the CREATE MCP SERVER privilege on the schema plus USAGE on the schema, the referenced Cortex Search service, SELECT on referenced semantic views, and USAGE on referenced warehouses, functions, or agents. At runtime, each tool acts with exactly the permissions of the calling role — an agent can only see and change what the connected role is already authorized to.

Difference from the earlier community project

Before the hosted server existed, an open-source community project at Snowflake-Labs/mcp (Apache-2.0) offered similar functionality and was installed locally via uvx or Docker. That repository now explicitly states it is no longer maintained and points to the officially managed server described here as the recommended successor. This entry covers only the current, hosted service; since that service is not itself published as open-source code, the entry carries no repository and no license field.

Cost

The documentation does not list a separate license fee for the MCP server itself; it is a feature of the Snowflake platform. Using it, however, incurs regular Snowflake costs: an existing Snowflake account, a warehouse for executing SQL and generic tools, and, depending on the tool types used, Cortex-specific consumption costs. Without an existing, paid Snowflake account, the server cannot meaningfully be used; concrete amounts are listed on Snowflake's official pricing page.

Who the server is for

The Snowflake-managed MCP server suits teams already working with Snowflake who want to make their data — structured tables via semantic models, unstructured documents via Cortex Search, or both combined through a Cortex Agent — accessible to AI assistants and agents, without operating their own bridging server or managing credentials outside Snowflake. For organizations without a Snowflake account, the server is naturally irrelevant; those looking for a lightweight, self-hosted SQL connection to a different data platform can find alternatives such as ClickHouse MCP Server or PlanetScale MCP Server elsewhere in this catalog.

Requirements

An existing Snowflake account with a role that has CREATE MCP SERVER on the target schema and access to the referenced objects (semantic views, Cortex Search services, warehouses, functions); an MCP-capable client that supports OAuth or bearer tokens.

Installation instructions

The Snowflake-managed MCP server is not a program to download but a database object inside a Snowflake account. This guide covers the basic steps from creation to a first connection with an AI client.

Quick start

Create the server with CREATE MCP SERVER <name> FROM SPECIFICATION $$ ... $$ in a schema, naming the desired tools (Cortex Analyst, Cortex Search, SQL execution, custom functions, Cortex Agent) in the YAML specification. In the client, configure the endpoint https://<account>.snowflakecomputing.com/api/v2/databases/<database>/schemas/<schema>/mcp-servers/<server_name> and authenticate via OAuth or a Programmatic Access Token.

Step 1: Prerequisites

You need an existing Snowflake account and a role with the CREATE MCP SERVER privilege on the target schema, plus USAGE privileges on the objects the server should expose later — for example a semantic view for Cortex Analyst, a Cortex Search service, or a warehouse for SQL execution.

Step 2: Create the server via SQL

The server is created with a YAML specification listing the available tools:

sql
CREATE OR REPLACE MCP SERVER my_server
FROM SPECIFICATION
$$
tools:
  - name: "query_metrics"
    identifier: "my_db.my_schema.my_semantic_view"
    type: "CORTEX_ANALYST_MESSAGE"
    description: "Answer questions about revenue and metrics in natural language."
    title: "Metrics"
$$;

Other tool types are CORTEX_SEARCH_SERVICE_QUERY for unstructured search, SYSTEM_EXECUTE_SQL for direct SQL execution, CORTEX_AGENT_RUN for an existing Cortex Agent, and GENERIC for custom functions or stored procedures. Per the documentation, a server allows up to 50 tools in total.

Step 3: Endpoint and authentication

The server is then reachable at https://<account>.snowflakecomputing.com/api/v2/databases/<database>/schemas/<schema>/mcp-servers/<server_name>. For production use, Snowflake's own OAuth 2.0 is the default path; for a quick first test, a Programmatic Access Token (PAT) used as a bearer token in the client works well. Use hyphens rather than underscores in hostnames to avoid connection issues.

Step 4: Connecting an MCP client

In a CLI client such as Claude Code, the server can be registered for testing with a command following this pattern, replacing the angle-bracketed placeholders with your own values:

Code
claude mcp add --transport http snowflake https://<account>.snowflakecomputing.com/api/v2/databases/<database>/schemas/<schema>/mcp-servers/<server_name>

When connecting, the client will ask for the bearer token or start the OAuth flow, depending on how the server is configured at the account level.

Keep permissions narrow

Create a dedicated, narrowly scoped role for MCP access instead of reusing an existing, broadly privileged role. This matters especially for the SYSTEM_EXECUTE_SQL tool, since an agent uses it to run model-generated SQL against real data. Before enabling it, check exactly which tables, semantic views, and search services the role can actually see.

Step 5: Verifying the setup

Start with a simple, read-only question in the client that matches the configured tool, such as a metrics question for Cortex Analyst. If the agent responds with a plausible result based on real data instead of a permission or connection error, the setup is correct.

Common pitfalls

A frequent mistake is using an underscore instead of a hyphen in the account identifier of the endpoint — this causes TLS or connection errors, since Snowflake consistently uses hyphens in hostnames. A second pitfall is missing privileges on referenced objects: if the semantic view or Cortex Search service lives in a different schema than the server, the role additionally needs USAGE on that schema, otherwise CREATE OR REPLACE MCP SERVER itself fails. If a tool returns an empty or implausible answer instead of an error, check the underlying semantic view or Cortex Search configuration first — the cause is often incomplete semantic modeling rather than the MCP connection itself. Anyone covering several use cases should also prefer several lean servers with clearly separated tools over one server mixing many different tool types, since that makes later permission management and troubleshooting noticeably easier.

claude mcp add --transport http snowflake https://<account>.snowflakecomputing.com/api/v2/databases/<database>/schemas/<schema>/mcp-servers/<server_name>

Authentication

Snowflake's own OAuth 2.0 by default, alternatively bound to an external identity provider (e.g. Okta, Microsoft Entra ID) at the account/database/schema level. For local development and testing, a Programmatic Access Token (PAT) used as a bearer token.

Required access permissions

Each tool acts with the permissions of the connected Snowflake role: SQL execution requires warehouse usage and the underlying table privileges, Cortex Analyst tools require read access to the referenced semantic view, and Cortex Search tools require usage rights on the search service. Roles should be scoped as narrowly as the use case allows.

Transmitted or stored data

Requests and model-generated SQL or search terms go to the Snowflake platform; responses such as query results, search hits, and agent outputs return to the MCP client and the connected AI provider. The exact data involved depends entirely on the objects wired into the server specification.

Security risks

A SYSTEM_EXECUTE_SQL tool lets an agent run model-generated SQL against real production data; without tightly scoped role privileges and review of sensitive objects, this can expose confidential data. Because all permissions flow through the connected role, MCP access should use a dedicated, narrowly scoped role rather than reusing a broadly privileged existing one. Cortex Search results drawn from document content can also contain embedded, misleading instructions (prompt injection).

License and costs

License
Not recorded yet.
Cost
paid

Per the documentation, the MCP server itself carries no separate license fee; using it requires an existing, paid Snowflake account and incurs regular warehouse and Cortex consumption costs. Concrete amounts are listed on Snowflake's official pricing page.

Alternatives

Not recorded yet.

At a glance

Provider
Snowflake
Status
Official server
Deployment
Remote
Current version
Not recorded yet.
Last reviewed
14.09.2026

Repository and documentation

Categories

Supported clients