Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@snowflake-mcp-server Query the sales table for total revenue by month for 2024"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Supercharge AI Agents with Native Snowflake Data Cloud & Cortex AI Superpowers! ⚡
An enterprise-grade Model Context Protocol (MCP) server providing 140 tools across 19 domain modules, dynamic profile switching, zero-config connection resolution, safe SQL execution, virtual warehouse management, object inspection, Horizon data lineage, and Cortex AI integrations straight to your favorite AI assistant.
flowchart TD
subgraph Clients["AI Clients & Hosts"]
Claude["Claude Desktop / Claude Code"]
Antigravity["Antigravity / Gemini CLI"]
CortexClient["Snowflake Cortex Agent"]
Cursor["Cursor / VS Code"]
end
subgraph Protocol["MCP Protocol Boundary (Spec 2026-07-28)"]
STDIO["stdio Transport"]
HTTP["Streamable HTTP Transport"]
end
subgraph Server["snowflake-mcp (FastMCP 4)"]
CLI["CLI & Arg Parser (Allowed Hosts & DNS Rebinding Protection)"]
Auth["Multi-Auth & Profile Resolver (~/.snowflake/connections.toml / Key-Pair / PAT / SSO)"]
Registry["Tool Registry (140 Tools across 19 Modules)"]
Safety["Safety Gates (confirm=True, Read-Only Guard, Row Limits)"]
end
subgraph Cloud["Snowflake Data Cloud"]
SQL["SQL & Transaction Engine"]
Warehouses["Virtual Warehouses & Scaling"]
Horizon["Horizon Lineage, Tags & Policies"]
Cortex["Cortex AI (Search, Complete, Analyst)"]
SPCS["SPCS Compute Pools & Services"]
Storage["Stages, Pipes, Streams & Iceberg"]
end
Clients --> STDIO & HTTP
STDIO & HTTP --> CLI
CLI --> Auth
Auth --> Registry
Registry --> Safety
Safety --> SQL & Warehouses & Horizon & Cortex & SPCS & StorageRelated MCP server: Snowflake MCP Server
Community Project Disclaimermcp-server-snowflake is an independent open-source project licensed under Apache 2.0. It is not affiliated with, sponsored by, endorsed by, or supported by Snowflake Inc. "Snowflake" and "Cortex" are trademarks of Snowflake Inc.
Safety Guardrails
Read-Only Safety Mode: Set SNOWFLAKE_MCP_READONLY=1, pass --readonly, or pass --profile readonly to block mutating tools. The profile sets the same read-only flag the gate reads. On each new session, read-only mode runs USE SECONDARY ROLES NONE before any tool SQL and closes the connection if that pin fails. Write mode does not run that statement. Objects readable only through a secondary role's grants are not visible under --readonly until the read-only primary role is granted SELECT on them directly. Set DEFAULT_SECONDARY_ROLES = () on the read-only user, or use a dedicated user that holds only that role. Prefer a programmatic access token with ROLE_RESTRICTION set to the read-only role so Snowflake evaluates privileges under that role (programmatic access tokens). ROLE_RESTRICTION does not replace the session pin. See FAQ / Troubleshooting.
Caller SQL: queries_query, queries_get_query_plan, and the query arguments of recipes_warehouse_scale_and_execute and recipes_export_query_to_stage refuse anything that is not one read-only statement, whether or not read-only mode is on. Allowed forms are one SELECT (no INTO), SHOW, DESCRIBE/DESC, or EXPLAIN SELECT. SYSTEM$ calls are refused except SYSTEM$TYPEOF and SYSTEM$CLUSTERING_INFORMATION. IDENTIFIER(...) in call position and TABLE(IDENTIFIER(...)) are refused. User-defined functions inside SELECT are not inspected.
Destructive Safety Gates: Dropping databases, schemas, or tables requires explicit confirm=True.
Object names follow Snowflake's identifier rules: unquoted names resolve to uppercase (mydb.public.orders is MYDB.PUBLIC.ORDERS), and a double-quoted part keeps its exact case and may contain . ("My.Db".PUBLIC."Orders"; write a literal " as ""). Pass an object created with a lowercase or mixed-case quoted name in double quotes. Every object row a SHOW-based list tool returns also carries quoted_name, its name already in double quotes with any " doubled (my_table gives "my_table"), ready to pass back as an object name argument; a row that has a database_name and a schema_name also carries quoted_qualified_name, the full "database"."schema"."name", which names the object from any session database and schema. An unterminated quote, an empty part, or a stray " in an unquoted part is refused before any SQL runs.
Name lookups through SHOW ... LIKE: warehouses_describe_warehouse, governance_describe_role and tags_describe_tag find their object with SHOW ... LIKE, which matches case-insensitively and treats _ and % as wildcards, with no documented escape (SHOW WAREHOUSES). A name in double quotes, such as a list row's quoted_name (or the tag part of quoted_qualified_name), is unquoted, sent to LIKE as is, and only the row whose name matches it exactly, case-sensitively, is returned; with no exact match the call fails with the usual not-found error rather than returning a near match. An unquoted name keeps the case-insensitive lookup. The SHOW WAREHOUSES LIKE fallback of warehouses_get_warehouse_load_history matches a quoted name the same way, and fails with the not-found error when no warehouse is left. Row cap: because _ and % widen the match, look-alike names count against the limits, in that fallback too, since it matches names the same way (a quoted name exactly and case-sensitively, an unquoted one case-insensitively): only the first SNOWFLAKE_MAX_ROWS rows (1,000 by default) are read, and SHOW returns at most 10,000 rows (SHOW WAREHOUSES; SHOW TAGS fails instead when there are more), so with enough look-alikes the target can be cut off and reported as not found. When no database resolves and the rows reach the row cap, a match in another database can be cut off too, so tags_describe_tag fails with the ambiguity error instead of returning the first row, even if every row it read comes from one database.
Query Limits: Default execution limits prevent context window overflow (SNOWFLAKE_MAX_ROWS=1000, SNOWFLAKE_QUERY_TIMEOUT=120).
In read-only mode (--readonly, --profile readonly, or SNOWFLAKE_MCP_READONLY=1), each new session runs USE SECONDARY ROLES NONE before any tool SQL. If that pin fails, the server fails closed: the connection is closed and not served. Write mode does not run the statement.
Objects readable only through a secondary role's grants need SELECT granted on the primary role (the read-only role) directly. Set DEFAULT_SECONDARY_ROLES = () on the read-only user, or use a dedicated user that holds only that role.
Some describe and lineage tools return an empty success for a missing name. That result is the tool behavior. The live harness records it as expect: success. An empty success here is not a harness bug:
warehouses_describe_warehouse
governance_describe_role
horizon_get_object_lineage
warehouses_describe_warehouse and governance_describe_role run SHOW ... LIKE and return status: success with empty details when an unquoted name matches nothing; a quoted name with no exact match is the not-found error. tags_describe_tag never returns an empty success: a tag with no match, quoted or not, is the not-found error. When no database resolves (for example schema.tag, or a bare tag with no database), SHOW TAGS LIKE runs with IN ACCOUNT — an explicit account scope, since without an IN clause the scope depends on the session: in a database context "Objects with the same name are only displayed once if no IN clause is used", and with no database in use it has "the same effect as specifying IN ACCOUNT" (SHOW TAGS) — and rows are filtered to the resolved schema client-side; with no database resolved it fails with an ambiguity error, asking for database.schema.tag, when the tag is found in more than one database or when SHOW TAGS returns as many rows as the row limit. horizon_get_object_lineage returns status: success when its OBJECT_DEPENDENCIES query succeeds, with empty upstream and downstream lists when the named object is absent.
horizon_get_column_lineage returns status: success when its SNOWFLAKE.ACCOUNT_USAGE.ACCESS_HISTORY query succeeds. That query does not filter on the supplied table or column name, so the payload can still contain recent access-history rows when that name is missing. The live harness still uses expect: success because the call succeeds. That is harness policy, not an empty-payload claim.
Most other missing-object describe tools still reject: the call is a tool error result (isError: true) whose text is the {"status": "error", "error": ...} JSON, and the live harness records expect: rejected. The fixture table and marker lists are in TESTING.md under Live e2e expect policy.
ROLE_RESTRICTIONPrefer ROLE_RESTRICTION on a programmatic access token set to the read-only role, so Snowflake evaluates privileges under that role (programmatic access tokens).
ROLE_RESTRICTION does not replace the session pin. Read-only mode still runs USE SECONDARY ROLES NONE on every new session. Live checks showed an unrestricted programmatic access token can still surface secondary roles until that statement runs.
mcp-server-snowflake automatically resolves credentials across all enterprise Snowflake configurations:
Snowflake CLI Inheritance (~/.snowflake/connections.toml):
Zero-configuration connection. If you have configured connections via snow, the server automatically connects to your default or specified profile (-c <conn_name>).
Dynamic Profile Switching:
Switch active connection profiles on the fly via governance_use_connection(connection_name) and inspect available profiles with governance_list_connections.
Programmatic Access Tokens (PAT) / OAuth:
Set token in connections profile or SNOWFLAKE_TOKEN.
RSA Key-Pair JWT Authentication:
Set SNOWFLAKE_PRIVATE_KEY_PATH (and optional SNOWFLAKE_PRIVATE_KEY_PASSPHRASE).
SSO / Browser Authentication:
Set SNOWFLAKE_AUTHENTICATOR=externalbrowser.
Environment Variables:
Standard SNOWFLAKE_ACCOUNT, SNOWFLAKE_USER, SNOWFLAKE_PASSWORD, SNOWFLAKE_WAREHOUSE, SNOWFLAKE_DATABASE, SNOWFLAKE_SCHEMA, SNOWFLAKE_ROLE.
# PyPI publication is pending dispute #11989.
# Install from GitHub
uvx --from git+https://github.com/christianclaudio/mcp-server-snowflake snowflake-mcp
# Or run the GHCR image
docker run -i --rm ghcr.io/christianclaudio/mcp-server-snowflake
# Interactive setup wizard
snowflake-mcp --init
# Run with a specific Snowflake CLI connection profile
snowflake-mcp -c my_connection
# Run in read-only mode (every tool stays registered; handlers reject mutations)
snowflake-mcp -c my_connection --readonly
# List one domain, or only read-only tools
snowflake-mcp --profile cortex
snowflake-mcp --profile readonly
# `--profile readonly` hides mutating tools, sets the read-only flag, and pins USE SECONDARY ROLES NONE on each session.
# Set DEFAULT_SECONDARY_ROLES = () on that user, or use a dedicated user that holds only the read-only role.
# Opt in to regex tool search instead of the flat 140-tool tools/list
snowflake-mcp --enable-tool-search
# Build a local image (the published image is ghcr.io/christianclaudio/mcp-server-snowflake)
docker build -t mcp-server-snowflake .
docker run -i --rm mcp-server-snowflakeDomain Suite | Count | Key Tools |
1. SQL Queries & Transactions | 9 |
|
2. Databases & Clones | 7 |
|
3. Schemas & Clones | 6 |
|
4. Tables, Views & Partitions | 10 |
|
5. Virtual Warehouses & Scaling | 8 |
|
6. Stages & File Operations | 6 |
|
7. Tasks & DAG Pipelines | 7 |
|
8. Streams & Change Data Capture | 5 |
|
9. Dynamic & Iceberg Tables | 9 |
|
10. Snowpipe & Ingestion | 5 |
|
11. Alerts & Notifications | 6 |
|
12. Governance, RBAC & Users | 12 |
|
13. Network & Password Policies | 6 |
|
14. SPCS Compute Pools & Streamlit | 8 |
|
15. Object Tags & Classifications | 4 |
|
16. Horizon Lineage & Governance | 6 |
|
17. Programmability, UDFs & Secrets | 10 |
|
18. Cortex AI & NLP Extensions | 8 |
|
19. Composite Agent Workflows | 8 |
|
mcp-server-snowflake works seamlessly with all major AI assistants, IDEs, and CLI tools via standard stdio or streamable-http.
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"snowflake": {
"command": "snowflake-mcp",
"args": ["-c", "my_connection"],
"env": {
"SNOWFLAKE_DEFAULT_CONNECTION_NAME": "my_connection"
}
}
}
}For Claude Code CLI:
claude mcp add snowflake -- snowflake-mcp -c my_connectionAdd to .agents/mcp_config.json (or global ~/.gemini/config/mcp_config.json):
{
"mcpServers": {
"snowflake": {
"command": "snowflake-mcp",
"args": ["-c", "my_connection"],
"env": {
"SNOWFLAKE_DEFAULT_CONNECTION_NAME": "my_connection"
}
}
}
}Add to ~/.snowflake/cortex/mcp.json:
{
"mcpServers": {
"snowflake": {
"command": "snowflake-mcp",
"args": ["-c", "my_connection"]
}
}
}Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"snowflake": {
"command": "snowflake-mcp",
"args": ["-c", "my_connection"]
}
}
}Add to cline_mcp_settings.json:
{
"mcpServers": {
"snowflake": {
"command": "snowflake-mcp",
"args": ["-c", "my_connection"]
}
}
}Launch snowflake-mcp as a long-running Streamable HTTP service:
snowflake-mcp --transport streamable-http --host 127.0.0.1 --port 8000Connect your HTTP client or proxy to endpoint http://127.0.0.1:8000/mcp.
HTTP authentication. Set SNOWFLAKE_MCP_AUTH_TOKEN to require Authorization: Bearer <token> on every HTTP request; a missing or wrong token gets 401. The token is stripped, and a blank value counts as unset. It is attached when the server is built, so snowflake-mcp, fastmcp run and an ASGI host mounting mcp.http_app() all enforce it when it is set. stdio never uses it.
With no token, snowflake-mcp still starts an HTTP bind to 127.0.0.1, ::1 or localhost, unauthenticated, and logs a warning. A tokenless bind to any other host exits with code 2. Set the token, bind to localhost, or set SNOWFLAKE_MCP_ALLOW_UNAUTHENTICATED_BIND to 1, true, yes or on to accept an unauthenticated public bind (any other value refuses). Other entry points get the token but not the localhost check: fastmcp run and http_app() do not go through snowflake-mcp's main(), so a bind to 0.0.0.0 with no token there is not refused and serves without authentication. The host or its process manager owns the bind address, so set SNOWFLAKE_MCP_AUTH_TOKEN there.
To serve the image over HTTP, pass credentials and the token from your environment or an env file, never on the command line:
# export SNOWFLAKE_ACCOUNT, SNOWFLAKE_USER, SNOWFLAKE_TOKEN and SNOWFLAKE_MCP_AUTH_TOKEN first, or use --env-file .env
docker run --rm -p 8000:8000 \
-e SNOWFLAKE_ACCOUNT -e SNOWFLAKE_USER -e SNOWFLAKE_TOKEN -e SNOWFLAKE_MCP_AUTH_TOKEN \
ghcr.io/christianclaudio/mcp-server-snowflake \
--transport streamable-http --host 0.0.0.0 --allowed-host mcp.example.comReplace mcp.example.com with the host name clients use to reach the server.
Known limits.
Only snowflake-mcp refuses a tokenless public bind (exit code 2). fastmcp run and mcp.http_app() enforce the token when it is set but do not refuse a tokenless public bind.
A token changed after the server is built is not picked up; restart the server.
In the default (stateful) HTTP mode, a session idle for 30 minutes expires: its next request gets HTTP 404 and the client must start a new session. Change this with FastMCP's session_idle_timeout (on http_app()) or FASTMCP_HTTP_SESSION_IDLE_TIMEOUT (seconds, or none to never expire); see the FastMCP 4.1.0 release and #5229.
All changes are strictly verified with automated safety and contract gates before release:
# 1. Format and lint checks
uv run ruff check .
uv run ruff format --check .
# 2. Strict static typing
uv run mypy src/
# 3. Unit and mocked test suite
uv run pytest
# 4. Tool contract verification (builds the server; 140 tools)
uv run python scripts/check_tool_contract.py
# 5. MCP protocol conformance suite (Spec 2026-07-28)
./scripts/check_conformance.sh
# 6. Release notes preview since the last tag (needs full history and tags)
python3 scripts/release_notes.pyApache 2.0. Copyright (c) 2026 Christian Claudio.
Create an alert with schedule, condition SQL, and action SQL. Destructive actions require confirm=True.
Inputs: schedule (required), action_sql (required), alert_name (required), condition_sql (required), +5 more. Returns: object
Describe alert condition SQL, action SQL, schedule, and warehouse.
Inputs: alert_name (required), database, schema_name. Returns: object
Drop an alert in Snowflake. Requires confirmation.
Inputs: alert_name (required), confirm, database, schema_name. Returns: object
List configured alerts with condition queries, schedules, and state. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Resume a suspended alert.
Inputs: alert_name (required), database, schema_name. Returns: object
Suspend an active alert.
Inputs: alert_name (required), database, schema_name. Returns: object
Describe compute pool instance family, min/max nodes, active nodes, and state.
Inputs: pool_name (required). Returns: object
Describe Streamlit application root location, stage, and query warehouse.
Inputs: streamlit_name (required), database, schema_name. Returns: object
List Snowpark Container Services (SPCS) compute pools. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern. Returns: object
List OCI image repositories in Snowflake. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List container services running on SPCS compute pools. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List Streamlit applications hosted in Snowflake. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Resume an idle or suspended SPCS compute pool.
Inputs: pool_name (required). Returns: object
Suspend an active SPCS compute pool to stop node provisioning costs.
Inputs: pool_name (required). Returns: object
Answer an analytical question via AI_COMPLETE. Optionally prefixes the prompt with semantic_model_path. Does not call Cortex Analyst REST (POST /api/v2/cortex/analyst/message).
Inputs: question (required), semantic_model_path. Returns: object
Run LLM completion via Snowflake AI_COMPLETE (e.g., 'llama3.3-70b', 'llama3.1-8b').
Inputs: prompt (required), model, max_tokens, temperature. Returns: object
Generate text embeddings via Snowflake AI_EMBED. Output dimension is set by the model (default snowflake-arctic-embed-m is 768); replaces legacy EMBED_TEXT_768.
Inputs: text (required), model. Returns: object
Extract answers from text via Snowflake AI_EXTRACT. Returns the AI_EXTRACT JSON object (response/error), not a legacy plain-string answer.
Inputs: question (required), source_text (required). Returns: object
Query a Snowflake Cortex Search Service index over unstructured text.
Inputs: query (required), service_name (required), limit, columns. Returns: object
Analyze sentiment via Snowflake AI_SENTIMENT. Returns an object with a categories array (overall plus optional aspects), not a legacy -1.0..1.0 float score.
Inputs: text (required). Returns: object
Summarize text via Snowflake AI_SUMMARIZE (drop-in for legacy CORTEX.SUMMARIZE).
Inputs: text (required). Returns: object
Translate text via Snowflake AI_TRANSLATE (drop-in for legacy CORTEX.TRANSLATE).
Inputs: text (required), source_language (required), target_language (required). Returns: object
Create a zero-copy clone of a database in Snowflake.
Inputs: source_database (required), target_database (required). Returns: object
Create a new database in Snowflake.
Inputs: name (required), comment, if_not_exists, data_retention_time_in_days. Returns: object
Describe properties, owner, retention time, and comment of a database.
Inputs: database_name (required). Returns: object
Drop a database in Snowflake. Requires confirmation flag.
Inputs: name (required), confirm. Returns: object
Retrieve the exact CREATE DATABASE DDL definition.
Inputs: database_name (required). Returns: object
List all available databases accessible to the current role. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern. Returns: object
Restore a recently dropped database using Time Travel.
Inputs: name (required). Returns: object
Describe dynamic table definition, target lag, warehouse, and query text.
Inputs: table_name (required), database, schema_name. Returns: object
Describe schema, catalog integration, and external volume of an Iceberg table.
Inputs: table_name (required), database, schema_name. Returns: object
List catalog integrations (Polaris, AWS Glue, Object Storage) configured in Snowflake. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern. Returns: object
List dynamic tables with lag targets, refresh mode, and last refresh status. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List external cloud storage volumes configured for Apache Iceberg tables. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern. Returns: object
List Apache Iceberg tables in the account, database, or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Trigger an immediate manual refresh of a dynamic table.
Inputs: table_name (required), database, schema_name. Returns: object
Resume scheduling and lag monitoring for a dynamic table.
Inputs: table_name (required), database, schema_name. Returns: object
Suspend automated refresh and lag evaluation for a dynamic table.
Inputs: table_name (required), database, schema_name. Returns: object
Create a new role in Snowflake.
Inputs: role_name (required), comment, if_not_exists. Returns: object
Create a new Snowflake user with default role and warehouse.
Inputs: user_name (required), comment, password, default_role, +2 more. Returns: object
Describe role properties and assigned grants. A name in double quotes, such as a list row's quoted_name, must match exactly.
Inputs: role_name (required). Returns: object
Describe user properties, email, disabled status, and default warehouse/role.
Inputs: user_name (required). Returns: object
Drop a role in Snowflake. Requires confirmation.
Inputs: role_name (required), confirm. Returns: object
Retrieve current active session context (current user, role, warehouse, database, schema, and account).
Inputs: no named parameters. Returns: object
List available connection profiles configured in ~/.snowflake/connections.toml and show active profile.
Inputs: no named parameters. Returns: object
List privileges granted to a specific role.
Inputs: role_name (required). Returns: object
List roles granted to a specific user.
Inputs: user_name (required). Returns: object
List roles available in the Snowflake account. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern. Returns: object
List users in the Snowflake account with login status and default roles. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern. Returns: object
Dynamically switch the active Snowflake session to a different connection profile from ~/.snowflake/connections.toml.
Inputs: connection_name (required). Returns: object
Describe signature, return type, and body of a masking policy.
Inputs: policy_name (required), database, schema_name. Returns: object
Describe signature, filter expression, and comment of a row access policy.
Inputs: policy_name (required), database, schema_name. Returns: object
Trace origin and transformation lineage for a column using Snowflake Horizon Access History.
Inputs: table_name (required), column_name (required), days, limit, +2 more. Returns: object
Retrieve upstream source and downstream dependent object lineage from Snowflake Horizon.
Inputs: object_name (required), database, direction, schema_name. Returns: object
List column masking policies in the account, database, or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List row access policies defined in the account, database, or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Describe allowed IP lists, blocked IP lists, and comments for a network policy.
Inputs: policy_name (required). Returns: object
Describe mode (INGRESS, INTERNAL_STAGE, EGRESS), type (IPV4, AWSVPCEID, HOST_PORT), and value list of a network rule.
Inputs: rule_name (required), database, schema_name. Returns: object
Describe password policy constraints (min length, lockout time, history, age).
Inputs: policy_name (required), database, schema_name. Returns: object
List network policies configured in the account. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern. Returns: object
List network rules defined in a database or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List password security policies defined in the account or database. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Create a Snowpipe for continuous ingestion.
Inputs: pipe_name (required), copy_statement (required), database, auto_ingest, +2 more. Returns: object
Describe pipe definition and COPY statement.
Inputs: pipe_name (required), database, schema_name. Returns: object
Drop a Snowpipe. Requires confirmation.
Inputs: pipe_name (required), confirm, database, schema_name. Returns: object
Get execution and health status of a Snowpipe (pending file count, last ingested timestamp).
Inputs: pipe_name (required), database, schema_name. Returns: object
List Snowpipes configured in a database or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Describe UDF function signature, return type, language, and body.
Inputs: function_signature (required), database, schema_name. Returns: object
Describe procedure signature, return type, language, and definition body.
Inputs: procedure_signature (required), database, schema_name. Returns: object
Describe secret metadata, type (GENERIC_STRING, OAUTH2), and owner without exposing secret value.
Inputs: secret_name (required), database, schema_name. Returns: object
List event tables used for application logging, tracing, and SPCS telemetry. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List user-defined functions (UDFs) in a database or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List external integrations (API, Storage, Notification, Security integrations). Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, integration_type. Returns: object
List notification integrations configured for alerts, tasks, and cloud messaging (SNS, PubSub, Webhooks). Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern. Returns: object
List stored procedures in a database or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List security secrets (API keys, OAuth credentials) stored in Snowflake. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List auto-increment sequences in a database or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Begin an explicit transaction on the active session.
Inputs: no named parameters. Returns: object
Cancel an active running Snowflake query by its Query ID. Requires confirmation.
Inputs: query_id (required), confirm. Returns: object
Commit the current active transaction on the session.
Inputs: no named parameters. Returns: object
Execute a data modification SQL statement (INSERT, UPDATE, DELETE, MERGE, CREATE). Requires confirmation.
Inputs: statement (required), confirm. Returns: object
Retrieve recent query execution history for the current account/user.
Inputs: limit. Returns: object
Retrieve operator-level execution statistics and profiling data for a past Query ID.
Inputs: query_id (required). Returns: object
Generate the EXPLAIN execution plan for a SQL query without executing it.
Inputs: query (required). Returns: object
Execute a SQL SELECT or read-only query on Snowflake and return structured rows with metadata.
Inputs: query (required), max_rows. Returns: object
Rollback the current active transaction on the session. Requires confirmation.
Inputs: confirm. Returns: object
Composite recipe: Summary of warehouse compute credits and storage consumption over the past 7 days.
Inputs: no named parameters. Returns: object
Composite recipe: Clone table with Time Travel timestamp or statement ID in 1 call.
Inputs: source_table (required), target_table (required), at_or_before. Returns: object
Composite recipe: Discover all tables and views in a schema along with their column definitions.
Inputs: database, schema_name. Returns: object
Composite recipe: Unload query results to a stage as Parquet or CSV files.
Inputs: query (required), stage_location (required), header, file_format. Returns: object
Composite recipe: Run health check on connection, active session, warehouse state, and credit consumption.
Inputs: no named parameters. Returns: object
Composite recipe: Describe table schema, row count, column types, and preview sample rows in 1 call.
Inputs: table_name (required), database, sample_rows, schema_name. Returns: object
Composite recipe: Profile table metadata, row count, and column inventory.
Inputs: table_name (required), database, schema_name. Returns: object
Composite recipe: Safely scale up a warehouse, run a heavy query, and optionally restore previous size. Requires confirmation.
Inputs: query (required), target_size (required), warehouse_name (required), confirm, +1 more. Returns: object
Create a zero-copy clone of a schema.
Inputs: source_schema (required), target_schema (required), database. Returns: object
Create a new schema inside a specified database.
Inputs: name (required), comment, database, if_not_exists. Returns: object
Describe schema properties, owner, and retention settings.
Inputs: schema_name (required), database. Returns: object
Drop a schema in Snowflake. Requires confirmation flag.
Inputs: name (required), confirm, database. Returns: object
List all schemas within a specific database or the current active database. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database. Returns: object
Restore a recently dropped schema using Time Travel.
Inputs: name (required), database. Returns: object
Create an internal named stage in Snowflake.
Inputs: stage_name (required), comment, database, schema_name, +1 more. Returns: object
Describe stage location URL, storage integration, and file format properties.
Inputs: stage_name (required), database, schema_name. Returns: object
Drop a stage in Snowflake. Requires confirmation.
Inputs: stage_name (required), confirm, database, schema_name. Returns: object
List files inside a Snowflake stage location (e.g. '@MY_STAGE/path/').
Inputs: stage_location (required), pattern. Returns: object
List internal and external stages available in the active database or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Remove a file from a stage location. Requires confirmation.
Inputs: stage_file_path (required), confirm. Returns: object
Create a CDC stream on a table or view.
Inputs: on_table (required), stream_name (required), database, append_only, +2 more. Returns: object
Describe stream metadata, source table, and stale status.
Inputs: stream_name (required), database, schema_name. Returns: object
Drop a stream. Requires confirmation.
Inputs: stream_name (required), confirm, database, schema_name. Returns: object
List table/view streams in a database or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Query unconsumed CDC changes recorded in a stream (default 10 rows).
Inputs: stream_name (required), limit. Returns: object
Create a zero-copy clone of an existing table.
Inputs: source_table (required), target_table (required). Returns: object
Create a table in Snowflake with specified column definitions SQL.
Inputs: table_name (required), columns_sql (required), database, schema_name, +1 more. Returns: object
Describe column definitions, data types, nullability, and primary keys for a table or view.
Inputs: table_name (required), database, schema_name. Returns: object
Drop a table in Snowflake. Requires confirmation flag.
Inputs: table_name (required), confirm. Returns: object
Retrieve the exact CREATE OR REPLACE TABLE/VIEW DDL definition for an object.
Inputs: object_name (required), object_type. Returns: object
List tables in a database schema with row counts and bytes. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
List views in a database schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Preview sample rows from a table (default 10 rows).
Inputs: table_name (required), sample_size. Returns: object
Truncate all data rows from a table while preserving schema. Requires confirmation.
Inputs: table_name (required), confirm. Returns: object
Restore a recently dropped table using Time Travel.
Inputs: table_name (required). Returns: object
Describe tag properties, allowed values, and comment. A name in double quotes, such as a list row's quoted_name, must match exactly.
Inputs: tag_name (required), database, schema_name. Returns: object
Get tag key/value assignments on a specific database object.
Inputs: object_name (required), object_domain. Returns: object
List object tags defined in a database or schema. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Set or assign a tag value on a Snowflake object (TABLE, SCHEMA, DATABASE, etc.).
Inputs: tag_name (required), tag_value (required), object_name (required), object_domain. Returns: object
Create a scheduled or serverless task in Snowflake.
Inputs: task_name (required), sql_statement (required), database, schedule, +3 more. Returns: object
Describe task definition, schedule, warehouse, and definition SQL.
Inputs: task_name (required), database, schema_name. Returns: object
Drop a task in Snowflake. Requires confirmation.
Inputs: task_name (required), confirm, database, schema_name. Returns: object
Trigger an immediate one-time execution of a task. Requires confirmation.
Inputs: task_name (required), confirm, database, schema_name. Returns: object
List tasks in a database or schema with schedule, state, and predecessor info. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern, database, schema_name. Returns: object
Resume a suspended task.
Inputs: task_name (required), database, schema_name. Returns: object
Suspend an active scheduled task.
Inputs: task_name (required), database, schema_name. Returns: object
Create a new virtual warehouse in Snowflake.
Inputs: warehouse_name (required), comment, auto_resume, auto_suspend, +2 more. Returns: object
Describe details and running state of a specific virtual warehouse. A name in double quotes, such as a list row's quoted_name, must match exactly.
Inputs: warehouse_name (required). Returns: object
Drop a virtual warehouse. Requires confirmation.
Inputs: warehouse_name (required), confirm. Returns: object
Get execution load, queuing, and provisioning history for a warehouse.
Inputs: warehouse_name (required). Returns: object
List virtual warehouses in the account with size, state, and auto-suspend configurations. Each row also has quoted_name, the name in double quotes, and, when the row has a database and schema, quoted_qualified_name.
Inputs: pattern. Returns: object
Change warehouse compute size (XSMALL, SMALL, MEDIUM, LARGE, XLARGE, 2XLARGE, etc.).
Inputs: size (required), warehouse_name (required). Returns: object
Resume a suspended virtual warehouse.
Inputs: warehouse_name (required). Returns: object
Suspend an active virtual warehouse to save compute costs.
Inputs: warehouse_name (required). Returns: object
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 140 tools
Most tools target a distinct resource and action, and category prefixes help separate domains. However, composite recipes duplicate primitive tools (e.g., recipes_inspect_table_with_sample vs tables_describe_table + tables_sample_table, recipes_clone_table_recipe vs tables_clone_table), and some describe/list tools overlap in grants coverage. Descriptions mitigate this, but agents still face non-trivial selection ambiguity.
Names generally follow a predictable snake_case pattern of category prefix + action + resource (databases_list_databases, warehouses_resize_warehouse, tasks_suspend_task). Minor deviations exist, such as cortex_search lacking an explicit verb/object structure, recipe names being phrase-like, and the 'streamlits' typo, but the convention remains readable and consistent.
At 140 tools, the surface is far beyond a well-scoped set and creates a heavy selection burden for an agent. Even for a broad platform like Snowflake, this is an extreme mismatch that encourages overlap, duplication, and cognitive overload.
Coverage is broad across databases, schemas, tables, warehouses, stages, tasks, streams, dynamic tables, governance, network, compute services, tags, Horizon, programmability, Cortex, and query execution. Some object types are list/describe-only (e.g., network policies, masking policies, functions/procedures/secrets, tag creation), but generic DML provides a workaround for most gaps.
Connect AI agents to 1000+ apps with managed authentication and tool-calling.
Connect your AI agent to 300+ methods and 100+ live tools
One place to build, share, and govern the skills and tools your AI agents use at work.
Connect AI agents to financial institution origination, analytics, and compliance workflows.
