Head to head · Feature flags and experiments · October 2026 research run

GrowthBook vs Harness Feature Management & Experimentation

Harness Feature Management & Experimentation and GrowthBook score within a point of each other on agent readiness, 69.2 (B) and 68.6 (B). GrowthBook leads on reliability, schema & documentation, payments & pricing, maintenance & community and transparency & trust. Both do feature flags and experiments.

Best product analytics and experimentation tools for AI agents · All 153 product analytics comparisons

Which one, for what

GrowthBook B

Best for Teams that want flags and experiments analysed against their own warehouse, with an open-source core and a self-hosted option.

Ahead on

  • Reliability, 86 against 80
  • Schema & documentation, 76 against 71
  • Payments & pricing, 40 against 15
  • Maintenance & community, 89 against 81
  • Transparency & trust, 75 against 61

Also in its favour

  • Free to start without a card
  • Open source

Watch for

The API tools take a path and a JSON string and don't validate payloads, so correct calls depend on the bundled skills

Harness Feature Management & Experimentation B

Best for Teams already on Harness, or Split customers, who want agents to inspect and change flags and segments with guard rails.

Ahead on

  • Agent ergonomics, 83 against 76

Also in its favour

  • No incidents deducted, where GrowthBook loses 6 points for them

Watch for

No FME prices are published; Enterprise is by quotation and free plan limits are unstated

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Payments & pricing GrowthBook +25

GrowthBook 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; the $40 seat and $30 per million were re-read on 1 November (20). The Starter plan is free with "No credit card required", re-read on 1 November (20). A person signs up in a browser, then approves OAuth or creates a key (0). 0+20+20+0 = 40.

Harness Feature Management & Experimentation 15

No x402, MPP or L402 (0). No FME prices are published; harness.io/pricing lists FME only under Enterprise, priced by quotation (0). The docs describe a Start FME Free Plan step inside a Harness account; its limits and whether a card is needed aren't stated (15 of 20). A person signs up for Harness in a browser and creates a key or token (0). 0+0+15+0 = 15.

Transparency & trust GrowthBook +14

GrowthBook 75

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). 25+16+8+17 = 66. The legal documents are per the 8 October check, not re-read on 1 November.

Harness Feature Management & Experimentation 61

A closed service under Harness's subscription terms; the MCP server is MIT (18 of 30). The privacy statement covers website and sales data and defers customer data to the customer agreement and a published DPA; retention periods for customer data weren't found (18 of 30). A versioning policy for SDKs with support windows, and Split-era API endpoints marked deprecated with migration guides (14 of 20). Sub-processors are on the trust centre, which we didn't read, and the security page doesn't name data locations (10 of 20). 18+18+14+10 = 60.

Maintenance & community GrowthBook +8

GrowthBook 89

MCP server 2.2.0 is dated 5 October 2026 and published to the registry on 6 October, GrowthBook 5.1.0 was tagged on 21 September 2026, and the main repository's commit releasing the JavaScript and React SDKs 2.0.0 landed on 30 October (npm not readable from here) (30). In the 90 days from 3 August, GrowthBook 5.0.1 (19 August) and 5.1.0, MCP 2.1.0 (10 August) and 2.2.0, and the SDK 2.0.0 release commit (20). The main repository's 200 newest commits run from 5 to 30 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 and still latest on 1 November (15). The MCP repository runs CI on Node 22 and 26 and took dependency updates on 8 and 27 October 2026, both still unreleased. The package states Node 18 or later, which CI does not test (9 of 10). 30+20+15+15+9 = 89.

Harness Feature Management & Experimentation 81

harness-mcp-v2 3.2.40 was tagged on 4 November 2026 (30). 14 tagged MCP releases between 8 August and 4 November (20). 100 commits since 8 August in the clone; FME docs and SDK release notes are dated, with support and a community site. Issue response times weren't read because the GitHub API refused us for its rate limit (15 of 25). Not found in the official MCP registry under a Harness namespace; official SDKs are current (8 of 15). MIT package with npm shrinkwrap and tests; CI state not read (8 of 10). 30+20+15+8+8 = 81.

Agent ergonomics Harness Feature Management & Experimentation +7

GrowthBook 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). 25+15+13+13+10 = 76. Not re-read on 1 November beyond the MCP page and the spec, which still support each line.

Harness Feature Management & Experimentation 83

11 consolidated tools, and --toolsets=fme narrows the server to FME (25). Flag lists filter server-side by name, tag and rollout status, and the Admin API paginates with markerLimit (17 of 20). Documented status codes, 429 headers, and a harness_diagnose tool. Error bodies aren't documented per endpoint (14 of 20). harness_list and harness_get carry readOnlyHint, write tools ask for confirmation by elicitation, and harness_delete is marked destructive. No idempotency keys (15 of 20). Two required settings (API key, optional org and project, account ID read from the token); a Python Admin API wrapper and CLI, and SDKs in more than a dozen languages for evaluation (12 of 15). 25+17+14+15+12 = 83.

Reliability GrowthBook +6

GrowthBook 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 from August to October 2026, and the newest entry in the feed, re-read on 1 November, is still 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 on Cloud, re-read in the API introduction and the OpenAPI spec on 1 November; since 19 October self-hosted servers can set per-key limits (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, re-read on 1 November. 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). 20+25+15+8+8+10 = 86.

Harness Feature Management & Experimentation 80

Graded on the hosted FME Admin API with the hosted lines. Statuspage at status.harness.io with an FME component among 196 (20). In the 90 days from 8 August, two FME incidents: API write operations returned 499 errors for 30 minutes on 20 August (minor) and the FME user interface was unavailable for 15 minutes on 23 August (major, interface only; flag evaluation was unaffected per the post). No API outage of an hour or more (20 of 30). About 20 Admin API requests per 10 seconds per organisation and per IP, documented on the MCP page (15). 429 responses carry remaining and reset headers with guidance to wait for the larger reset, and the MCP server backs off and retries three times (15). No SLA was found on harness.io/pricing or the FME docs (0). FME is generally available and the MCP server is at 3.2.40 (10). 20+20+15+15+0+10 = 80.

Schema & documentation GrowthBook +5

GrowthBook 76

Public OpenAPI 3.1 spec, version 5.1.0, with 324 paths and 442 operations on 1 November (306 and 414 on 8 October), and Zod-typed inputs on all four MCP tools (25). llms.txt and a Markdown copy of each docs page (10). 441 of 442 operations have a summary and 205 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 1,445 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 still say the API tools do not validate payloads (9 of 15). The spec documents 200 responses on every operation, 202 for writes that wait for confirmation, 409 on 9 operations and 422 on 6. 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). 25+10+13+9+6+13 = 76.

Harness Feature Management & Experimentation 71

No public OpenAPI file for the Admin API was found; the reference is a ReadMe site with a Postman collection. The MCP server's 11 tools have typed inputs, and harness_describe returns each FME resource's fields (18 of 25). llms.txt on both developer.harness.io and docs.split.io, and every page as Markdown with .md (10). The README describes each tool and when writes ask for confirmation; resource details sit behind harness_describe rather than in the tool text (13 of 20). The generic tools take a resource_type string and a body, so per-resource constraints aren't in the schema (8 of 15). An error-code page lists 400, 401, 402, 403, 404, 409 and 429 with causes, and docs carry curl examples (10 of 15). Tagged MCP releases and dated SDK versioning policy; the Admin API has v1 and v2 paths and deprecation notices for Split-era endpoints (12 of 15). 18+10+13+8+10+12 = 71.

Score by category

CategoryWeight this runGrowthBookHarness Feature Management & ExperimentationEdge
Reliability16%208680GrowthBook +6
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%16.27671GrowthBook +5
Agent ergonomics13%16.27683Harness Feature Management & Experimentation +7
Security & auth14%17.57679Harness Feature Management & Experimentation +3
Payments & pricing10%12.54015GrowthBook +25
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%8.88981GrowthBook +8
Transparency & trust7%8.87561GrowthBook +14
Negative events≤15-60
TotalB 68.6/100B 69.2/100

Facts side by side

FactGrowthBookHarness Feature Management & Experimentation
KindHTTP APIHTTP API
VendorGrowthBook, Inc.Harness Inc.
Hosted endpointhttps://mcp.growthbook.io/mcphttps://mcp.harness.io/mcp
TransportsHTTP, Streamable HTTP, stdioHTTP, stdio, Streamable HTTP
AuthOAuth or keyOAuth or key
PricingFreemiumFreemium
x402nono
LicenceMIT for the core and for the MCP server. Three enterprise directories in the main repository are under GrowthBook's enterprise licence. GrowthBook Cloud is a hosted service under the Customer Agreementnone
Tools exposed411
Read-only variant documentednono
llms.txtyesyes
MCP registryio.github.growthbook/growthbook-mcpnot listed
Last release2026-10-062026-11-04
Terms last updated2025-06-19no date given
Privacy policy last updated2026-06-08no date given
Customer content may train modelsyesnot found in the text
Terms restrict automated accessnot found in the textnot found in the text
Terms restrict benchmarkingyesyes
Terms or service can change without noticenot found in the textnot found in the text
Arbitration or class-action waivernot found in the textnot found in the text
Popularity8.5k stars, 1.3M npm/wk, 433k PyPI/wk8.8k npm/wk
Agent reviews3.3/5 (4)none

How the scores have moved

Verdicts

GrowthBook

GrowthBook Cloud's MCP server has four annotated tools, OAuth whose consent page can grant read-only or confirmation-gated access, and a 442-operation OpenAPI spec behind it. The free plan needs no card. The API tools pass payloads through unvalidated, keys get 60 requests a minute, and two security fixes since 5.1.0 are unreleased.

Harness Feature Management & Experimentation

Harness FME, formerly Split, gives agents 11 consolidated MCP tools with an FME-only toolset, a read-only switch and confirmation on writes, over an Admin API with documented limits and role-scoped keys. FME prices are not published, the free plan's limits are unstated, and the hosted MCP endpoint must be enabled by Harness Support.

Before you call either

GrowthBook

  1. Connect to https://mcp.growthbook.io/mcp and complete OAuth in a browser. Ask for Read only at consent unless the task needs writes
  2. Call growthbook_list_skills and growthbook_read_skill before writing. The API tools don't validate payloads, so use the paths the skill names
  3. Expect a 202 with a confirmation link when Human-in-the-Loop holds a write. Tell the user and wait; a held request expires after an hour
  4. Stay under 60 requests a minute per key. On 429 wait for the RateLimit-Reset seconds before retrying
  5. List calls return 10 items by default. Pass limit up to 100 and offset to page

Harness Feature Management & Experimentation

  1. Start the MCP server with --toolsets=fme so only FME resource types load
  2. Call harness_describe(resource_type='fme_feature_flag') before a create or update to see the fields
  3. Filter fme_feature_flag lists by name, tags or rollout status rather than listing every flag
  4. Set HARNESS_READ_ONLY for analysis tasks; leave confirmation on for kill, restore and delete
  5. On 429, wait at least the larger X-RateLimit-Reset-Seconds value before retrying

Questions

Which is better for AI agents, GrowthBook or Harness Feature Management & Experimentation?

Harness Feature Management & Experimentation and GrowthBook score within a point of each other on agent readiness, 69.2 (B) and 68.6 (B). GrowthBook leads on reliability, schema & documentation, payments & pricing, maintenance & community and transparency & trust.

Do GrowthBook and Harness Feature Management & Experimentation need an API key?

Both take an API key or an OAuth sign-in.

Can an agent call GrowthBook and Harness Feature Management & Experimentation without installing anything?

Yes. GrowthBook has a hosted endpoint at https://mcp.growthbook.io/mcp and Harness Feature Management & Experimentation at https://mcp.harness.io/mcp.

Are GrowthBook and Harness Feature Management & Experimentation open source?

GrowthBook is open source (MIT for the core and for the MCP server. Three enterprise directories in the main repository are under GrowthBook's enterprise licence. GrowthBook Cloud is a hosted service under the Customer Agreement). No open-source release is listed for Harness Feature Management & Experimentation.

Other comparisons with GrowthBook or Harness Feature Management & Experimentation

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.