Tools

Tools offered by the StitchDB MCP server. Read-only connections do not see tools that change anything.

#Projects

Tool	What it does
`list_projects`	Projects the connection can access: reference, name, status, region, API URL.
`get_project`	Status, region, usage, plan limits and the database host (never the password).
`get_project_url`	The API URL of a project.
`get_anon_key`	The public anon key. Secret keys are never returned.

#Database

Tool	What it does
`list_tables`	Tables and views with Row Level Security status, policy count, estimated rows, size and API exposure.
`describe_table`	Columns, keys, foreign keys, indexes, triggers and policies of a table.
`list_extensions`	Available and installed Postgres extensions.
`list_policies`	Row Level Security policies in `public` and `storage`.
`execute_sql`	Runs SQL as the project owner role: 15 second timeout, up to 1,000 rows. On read-only connections it runs one statement in a `READ ONLY` transaction.
`apply_migration`	Read-write only. Applies a named schema migration in one transaction and records it.
`list_migrations`	Migrations applied with `apply_migration`.
`generate_typescript_types`	A `Database` type for stitchdb-js from your tables, views, functions and enums.

#Operations

Tool	What it does
`get_logs`	Logs of API requests (`api`), Auth (`auth`), the REST server (`rest`), Postgres (`postgres`), Edge Functions (`functions`) or the SQL history (`sql`), with search, level and status filters, up to your plan's log retention.
`query_logs`	Counts and statistics over the logs of up to 24 hours: `SELECT status, count(*) AS n, p95(ms) FROM api WHERE status >= 500 GROUP BY status ORDER BY n DESC`. Sources `api`, `auth`, `rest`, `postgres`, `functions`; filters, group by (also by `minute`/`hour`), count, sum, avg, min, max and p50–p99. Read-only; log lines are returned as data.
`get_advisors`	Security, performance and health findings (`type`: `security`, `performance`, `health` or `all`). Health: error rates above normal for REST, Auth, Storage and Functions, failed sign-ins, connection and pooler pressure. Security and performance, with SQL to fix them: tables without RLS, RLS without policies, views that bypass RLS, exposed sensitive columns, missing primary keys, unindexed foreign keys and functions with a mutable search_path.
`list_connections`	Sessions connected to the database (as Database › Connections in Studio): state, run times, lock waits and the sessions blocking them, long-running and idle-in-transaction flags, and your roles' connections against their limit. Platform services' SQL is hidden. Works on read-only connections; cancelling or closing a session is done in Studio.
`list_storage_buckets`	Buckets with visibility, limits, object count and size.

#Edge Functions

Tool	What it does
`list_edge_functions`	Functions with URL, version and whether a JWT is required.
`get_edge_function`	Source files and settings of the live (or a given) version.
`deploy_edge_function`	Read-write only. Creates a function or deploys a new version from source files. See Edge Functions.
`get_edge_function_logs`	Recent console output, boot errors and limit stops.

#Provisioning and operating projects

For platforms and agents that create projects on the fly (see Platforms). Write tools need a read-write connection and, with an organization token, the matching scope.

Tool	What it does
`create_project`	Creates a project (optional labels, `ttl_seconds` for automatic deletion, `template_sql` and `seed_sql`). Returns at once; poll `get_project` until it is active.
`pause_project`, `restore_project`	Pause (data kept, API answers 503) and restore.
`set_project_limits`	Per-project caps below the plan (size, requests per second, end users, functions, …).
`rotate_api_keys`	New anon and service_role keys, old ones refused after a grace period.
`get_usage`, `get_organization_usage`	Daily usage of a project / of every project of the organization.
`list_webhooks`	The organization’s webhook endpoints.

#Docs and search

Tool	What it does
`search_docs`	Searches this documentation.
`search`, `fetch`	Search docs and your tables, then fetch one result (the pair ChatGPT uses for connectors and deep research).

Every tool carries readOnlyHint / destructiveHint annotations, so clients can ask before running anything that changes data.