GrowthBook

# Fix list: GrowthBook

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/growthbook, the October 2026 research run, assessed 8 October 2026. Grade BB, 70.1 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on GrowthBook: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No x402, MPP or L402 (0). Prices are public without a login. Pro is $40 a seat a month, with $10 per million CDN requests, $1 per GB and $30 per million Managed Warehouse events above the allowance (20). The Starter plan is free with no card (20). A person signs up in a browser, then approves OAuth or creates a key (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 2. Security & auth, 68 out of 100, up to 5.6 more on the total

Why it scored 68: OAuth with PKCE, dynamic client registration, refresh tokens and a revocation endpoint for the hosted MCP server. Its only scopes are openid, profile, email and offline_access, so a token carries the user's full permissions. Secret Keys take any built-in or custom role with per-project overrides, environment limits and expiry, and Personal Access Tokens can be limited. Secrets travel in headers only (26 of 30). A Read Only role, an endpoint with the API tools alone, and read and write tools with annotations. Draft, review and publish steps live in the skills, approval workflows are an Enterprise feature, and since 2.1.0 confirmation is left to the client. Free plan members can only hold the Admin role (14 of 20). No guidance on prompt injection was found, though the tools return names and descriptions that people wrote (3 of 15). Audit logging is an Enterprise feature with export. Personal tokens attribute actions to a user, and the MCP server labels its API calls with `X-GB-MCP-*` headers (10 of 15). SOC 2 Type II, a disclosure process with a form and SECURITY.md, routine penetration testing and public GitHub advisories, the latest on 19 August 2026. Rewards are merchandise only and there is no security.txt (15 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 3. Schema & documentation, 76 out of 100, up to 3.9 more on the total

Why it scored 76: Public OpenAPI 3.1 spec, version 5.1.0, with 306 paths and 414 operations, and Zod-typed inputs on all four MCP tools (25). llms.txt and a Markdown copy of each docs page (10). 413 of 414 operations have a summary and 185 have a description. The MCP tool descriptions say when to use each tool, and the bundled skills carry the workflows (13 of 20). The spec has 1,404 enums and marks required fields, but `growthbook_api_write` takes the path as a string and the body as a JSON string, and the docs say the API tools do not validate payloads (9 of 15). The spec has 17 examples and documents only 200 responses, plus 409 on 9 operations. The introduction lists eight status codes and a `message` body (6 of 15). Path versions /v1 and /v2, 34 operations marked deprecated, GitHub releases, and an MCP changelog with breaking changes called out (13 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 4. Agent ergonomics, 76 out of 100, up to 3.9 more on the total

Why it scored 76: Four compact MCP tools, with skills loaded on demand and a /mcp/api endpoint that drops the skill tools (25). `limit` (default 10, maximum 100) and `offset` on 40 list operations, with filters by project, data source and status. No field selection was found, and the API tools return the raw response body (15 of 20). Errors are a status code and a `message` with no machine code. 422 marks a soft warning that can be resubmitted with `ignoreWarnings`, and the MCP server adds recovery hints for 401, 403, 404 and 429 (13 of 20). All four tools carry readOnlyHint and destructiveHint, and reads and writes are separate tools. Every write is marked destructive, and no idempotency keys were found (13 of 20). The API tools need one to three parameters. The SDKs cover flag evaluation in many languages, not the REST API (10 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 5. Reliability, 86 out of 100, up to 2.8 more on the total

Why it scored 86: Graded on GrowthBook Cloud with the hosted lines. Instatus page at status.growthbook.io with three components (GrowthBook Cloud App, Api, CDN) and a notice history (20). No notices for August, September or October 2026, and the newest entry in the feed is dated 21 August 2025. The MCP server is not a listed component, and the 2.2.0 changelog fixes intermittent 502s in HTTP mode that were not posted, so 25 of 30. API keys are limited to 60 requests a minute (15). The docs say only to try again later on 429. The server source sends standard RateLimit headers and the MCP server waits for `RateLimit-Reset` and retries three times, but no backoff guidance or idempotency keys were found in the docs (8 of 15). The pricing page lists a 99.99 per cent uptime SLA on Enterprise. The SLA document itself was not found in public (8 of 10). The REST API's /v1 is described as stable and the MCP docs carry no beta label (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 6. Transparency & trust, 74 out of 100, up to 2.3 more on the total

Made of editorial 66, provenance 81.

Why it scored 74: The core and the MCP server are MIT. Three enterprise directories in the main repository are under GrowthBook's enterprise licence and are part of what Cloud runs (25 of 30). A Customer Agreement, a Privacy Notice of 8 June 2026 and a DPA of 19 June 2025 are published. The agreement grants GrowthBook an unrestricted right to use Customer Data, including to train its AI algorithms, then limits personal information to performing the Services. The Privacy Notice gives no retention periods, and the docs say only aggregate results are stored. We did not read the DPA, which is a download (16 of 30). Legacy v1 routes and 34 operations are marked deprecated with no removal dates, and no deprecation policy was found. The MCP changelog gives migration steps (8 of 20). The sub-processor list of 11 May 2026 names seven providers with locations, mostly in the United States. Notice of new ones is promised only as far as the law requires (17 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: growthbook.io, registered 2020-05-19 (6 years) (11 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects, and has 2 clauses that cost points (5.1 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 89 out of 100, up to 1 more on the total

Why it scored 89: MCP server 2.2.0 is dated 5 October 2026 and GrowthBook 5.1.0 was tagged on 21 September 2026 (30). Six releases since 10 July 2026, three of the app and three of the MCP server (20). The main repository's 200 newest commits run from 15 September to 8 October 2026, and there is a community Slack. We did not read how quickly issues are answered (15 of 25). `io.github.growthbook/growthbook-mcp` is in the official MCP registry at 2.2.0, published 6 October 2026 (15). The MCP repository runs CI on Node 22 and 26 with 34 tests and took dependency security updates on 8 October 2026. The package states Node 18 or later, which CI does not test (9 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- 10 August 2026. MCP server 2.1.0 replaced `growthbook_call_api` with two tools in a minor version, eleven days after 2.0.0 removed every 1.x tool, while the docs tell users to install `@growthbook/mcp@latest`. Both are marked breaking in the changelog with migration steps, and we found no advance notice, so the deduction is the minimum (https://github.com/growthbook/growthbook-mcp/blob/main/CHANGELOG.md).

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: how quickly GitHub issues and pull requests are answered. The GitHub API refused us for its rate limit and we did not read the issue lists.
- unchecked: the DPA of 19 June 2025, which is a download, and the contents of trust.growthbook.io, which is drawn by script.
- unchecked: the live tool list from https://mcp.growthbook.io/mcp, which needs an OAuth sign-in. Tool definitions were read from the 2.2.0 source.
- The text of the Enterprise SLA was not found in public. The 99.99 per cent figure is from the pricing and security pages.
- The docs page says the server has four tools, while the 2.0.0 changelog describes three. The count of four is from the 2.2.0 source.
- The compliance docs mention an active bug bounty programme, while the disclosure page says no financial rewards are paid.
- Whether the hosted server's tool changes of 30 July and 10 August 2026 were announced in advance was not established.
- The official registry entry lists only the npm stdio package and no remote URL.
- The authentication text inside the OpenAPI spec still says Secret Keys are admin or readonly, while the docs page says any role.

## Weaknesses

- The API tools take a path and a JSON string and don't validate payloads, so correct calls depend on the bundled skills
- API keys are limited to 60 requests a minute on Cloud, and no idempotency keys were found
- Audit logging is an Enterprise feature, and Free plan members can only hold the Admin role
- The Customer Agreement grants GrowthBook a right to use Customer Data to train its AI algorithms, and no opt-out was found
- MCP 2.0.0 and 2.1.0 replaced the whole tool surface in July and August 2026, the second in a minor version

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Connect to https://mcp.growthbook.io/mcp and complete OAuth in a browser. Use https://mcp.growthbook.io/mcp/api for the two API tools alone
- Call `growthbook_list_skills` and `growthbook_read_skill` before writing. The API tools don't validate payloads, so use the paths the skill names
- Stay under 60 requests a minute per key. On 429 wait for the `RateLimit-Reset` seconds before retrying
- List calls return 10 items by default. Pass `limit` up to 100 and `offset` to page
- A 422 is a soft warning. Resubmit with `?ignoreWarnings=true` only after reading the message

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.