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 fieldGrowthBookWatchlist candidateDocker MCP GatewayMCP100 Selected
Task fit
Best for
  • GrowthBook teams that want a small MCP surface for skills plus authenticated GrowthBook API operations.
  • Running multiple MCP servers behind one stdio connection
  • Isolating catalog MCP servers in containers with restricted privileges and resources
  • Centralizing server profiles, secrets, OAuth, logs, and call traces
Not ideal for
  • Users who expect a large purpose-built tool set or a hard technical confirmation gate before every mutation.
  • Users who need only one simple remote MCP endpoint
  • Hosts where Docker Desktop or Docker Engine cannot run
Avoid when
  • Do not rely on prompt/instruction-level confirmation guidance as a hard security boundary for POST/PUT/PATCH/DELETE API calls.
  • You cannot review every downstream MCP server and profile permission
  • The agent must not be able to discover or add tools dynamically
  • Container isolation is being treated as protection from an over-privileged downstream credential
Provenance
Provenance details

First-party MCP

io.github.growthbook · Publisher source ↗

First-party MCP

Docker · Publisher source ↗

Maintenance
Maintenance details

Stale

Repo: Not documented

Package: Not documented

Repo: Aug 13, 2026

Package: Not documented

Popularity evidence
GitHub starsNot documented

1,529

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

Client coverage
Client coverage detailsNot documented
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
AuthenticationHTTP mode supports OAuth; stdio mode can use a GrowthBook API key/personal access token.The gateway handles downstream server credentials and OAuth flows. The client connects locally over stdio by running `docker mcp gateway run`, optionally with a named profile.
CostGrowthBook plan/API usage and model/provider costs apply.The open-source gateway documents no standalone usage fee. Docker Desktop licensing, Docker AI Governance invite-only features, container infrastructure, downstream MCP services, and AI-client charges may apply.
PermissionsThe generic API call can reach operations allowed by the credential, including mutations. Publisher guidance asks agents to confirm mutations, but this is not a hard security gate.The gateway exposes the servers and tools included in the selected profile. It starts catalog servers in containers with restricted privileges, network access, and resource usage, injects configured secrets, and can support OAuth. Dynamic MCP can additionally expose tools that discover, add, configure, remove, and execute servers during a session.
Data handlingGrowthBook feature flag/experiment/project data and API payloads can flow to the AI client/provider.Tool requests and results pass through the local gateway to downstream MCP containers or remote services. Docker documents built-in logging and call tracing. Credentials are managed and injected by the gateway; downstream servers can still receive data and reach networks allowed by their configuration.
Limitations
Tradeoffs
  • Promoted after strict re-audit: canonical identity and MCP-native/publisher-backed implementation are sufficiently evidenced, with real workflow value, usable documentation, understandable permissions, plausible maintenance, and no unresolved security veto. Still not strong enough for Selected.
  • Gateway isolation improves lifecycle and credential management but adds Docker runtime complexity and local resource use.
  • The MCP Toolkit is documented as Beta, and Dynamic MCP is experimental.
  • The gateway reduces direct host privileges, but network access and injected credentials still let downstream servers affect external systems.
Risk contextBroad API passthrough under one tool can make mutating authority less obvious than a purpose-specific tool inventory.Gateway safety is only as strong as the selected profile, catalog provenance, tool filters, network policy, and downstream credentials. Disable experimental Dynamic MCP unless required, use narrow profiles, verify signatures where available, inspect traces, and treat each downstream server as a separate trust decision. This review does not claim MCP100 execution testing.
Evidence date
Editorial review2026-09-13T00:00:00Z2026-07-24
Candidate evidence2026-09-13T00:00:00Z2026-08-14T18:46:26.000Z

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

Read the evidence method →
Compare MCPs | MCP100 Index