Field-by-field comparison

Compare the evidence before choosing an MCP.

Each signal stays separate, missing facts remain visible, and the column order follows your selection.

MCPs in this comparison

Uncheck an MCP to remove it. Comparing two or three keeps each signal in its own row.

Open canonical comparison link

Comparable alternatives

Shown only when capability or industry context overlaps.

Side-by-side MCP evidence comparison
Evidence fieldSupabase MCPMCP100 SelectedGoogle MCP Toolbox for DatabasesMCP100 Selected
Task fit
Best for
  • Inspecting one Supabase project's schema, logs, advisors, and docs
  • Development environments using project-scoped read-only access
  • Teams that understand database and deployment consequences
  • Schema exploration and database-aware coding in development environments
  • Connecting one MCP layer to supported Google Cloud and third-party databases
  • Defining structured, parameterized database tools with integrated authentication and observability
Not ideal for
  • Production database administration by an unattended agent
  • Users needing only public Supabase documentation
  • Organizations unable to grant Supabase OAuth access to an MCP client
  • Unrestricted natural-language SQL against production data
  • Teams unwilling to operate and secure a local or hosted Toolbox process
Avoid when
  • The server would be left organization-wide rather than project-scoped
  • Write-capable SQL, migrations, functions, branches, or storage changes cannot be human-approved
  • Production secrets or customer records could be returned to the model
  • The configured database user can modify production data or schema beyond the approved task
  • Sensitive query results cannot be sent to the AI client
  • Credentials would be committed in tools.yaml or MCP configuration
Provenance
Provenance details

First-party MCP

Supabase · Publisher source ↗

First-party MCP

Google · Publisher source ↗

Maintenance
Maintenance details

Repo: Aug 14, 2026

Package: Not documented

Repo: Aug 14, 2026

Package: Aug 14, 2026

Popularity evidence
GitHub stars

2,863

GitHub stars · checked 2026-08-14T18:46:26.000Z

16,173

GitHub stars · checked 2026-08-14T18:46:26.000Z

30-day package downloadsNo reliable download data

61,660

30-day package downloads · checked 2026-08-14T18:46:26.000Z

External adoption evidence
  • github: repository stars: 2,863Exact attributable signal.github: repository stars · Admission snapshot · checked 2026-08-14T18:46:26.000ZEvidence source ↗
  • github: repository stars: 16,173Exact attributable signal.github: repository stars · Admission snapshot · checked 2026-08-14T18:46:26.000ZEvidence source ↗
Client coverage
Client coverage details
Claude Code✓ Streamable HTTP
Claude Desktop✓ Streamable HTTP
Codex✓ Streamable HTTP
VS Code✓ Streamable HTTP
Cursor✓ Streamable HTTP
OpenCode✓ Streamable HTTP
Claude Code✓ Local stdio
Claude Desktop✓ Local stdio
Codex✓ Local stdio
VS Code✓ Local stdio
Cursor✓ Local stdio
OpenCode✓ Local stdio
Client coverage noteLocal stdio runs on your computer. Streamable HTTP connects to a remotely hosted MCP server.
Access and data
AuthenticationHosted MCP uses browser OAuth and grants organization access to the MCP client. Project scope, read-only mode, and feature groups are encoded in URL query parameters.Authentication depends on the database source and can include database username/password, Google Cloud IAM, or other source-specific mechanisms. Use a dedicated read-only identity and secret manager or protected environment variables.
CostThe MCP server is open source; normal Supabase plan, database, branch, Edge Function, storage, and usage charges apply. Branching and some storage operations require a paid plan.The Apache-licensed Toolbox and npm server have no documented MCP usage fee. Database services, Google Cloud resources, network transfer, observability backends, hosting, and the AI client may incur normal charges.
PermissionsWithout project_ref the hosted server can access all projects in the authorized organization. With project_ref it is limited to one project. read_only=true uses a read-only Postgres user for SQL and disables listed mutation tools. Feature groups can further narrow exposed tools.Effective permissions come from each configured database identity and the enabled prebuilt or custom tools. Generic prebuilt sets can include schema discovery and execute_sql; custom tools can constrain statements, parameters, and accessible sources.
Data handlingEnabled tools may return schemas, table data from executed queries, logs, security/performance advisors, project metadata, publishable keys, types, and Edge Function source. Hosted mode sends tool requests through Supabase's MCP service.The Toolbox process connects directly to configured databases and returns selected schemas, rows, query results, and errors to the MCP client. Credentials may be supplied through environment variables or configuration; integrated cloud authentication and OpenTelemetry are available.
Limitations
Tradeoffs
  • Project scoping removes account-level tools but materially reduces blast radius.
  • Read-only mode disables mutations but also removes migration, deployment, branch, and storage configuration workflows.
  • Branching and some storage capabilities require paid plans; the server is pre-1.0 and may change.
  • Prebuilt tools are quick to adopt but may expose generic execute_sql capability; custom tools require more design but provide a narrower contract.
  • The npx path is convenient but the publisher recommends binary or container installation for standard reliable deployments.
  • Supporting many database engines increases configuration flexibility and operational complexity.
Risk contextUnscoped or write-enabled access can affect every project in an organization, execute SQL, apply migrations, deploy functions, or manage branches. Always set project_ref and read_only=true first, minimize feature groups, and avoid production data.Generic SQL tools can expose or mutate large amounts of data and retrieved database text can carry prompt-injection content. Prefer a non-production replica, read-only credentials, allow-listed custom parameterized tools, query limits, audit logs, and no secrets in config. This review does not claim MCP100 execution testing.
Evidence date
Editorial review2026-07-242026-07-24
Candidate evidence2026-08-14T18:46:26.000Z2026-08-14T18:46:26.000Z

Popularity, maintenance, fit, permissions, and client support are independent evidence fields.

Read the evidence method →