Niches · Product analytics & experimentation
Create, target or roll out a feature flag or experiment
One company's docs show a full fit.
What counts as done
The flag or experiment exists with precisely the requested targeting and split, other fields unchanged, and (for experiments) a running state with a start date, confirmed by a read-back.
Measured, for our own tests: this takes at most 4 calls. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: create a flag for the new checkout
, turn this flag on for users in Canada
, roll out to 20 per cent
, launch an experiment on iOS and Android
, add these emails to the flag
Does the job 1
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| LaunchDarkly BB Catamorphic, Co. (dba LaunchDarkly) | create-flag, update-flag-targeting, update-targeting-rules and update-rollout take semantic-patch instructions that change only what is named, and start-experiment-iteration starts an experiment; approvals apply where an environment requires them. source |
Does the job, with a limit 11
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Amplitude BB Amplitude, Inc. | Management API PATCH takes optional fields including startDate and decision, but targetSegments replaces existing segments and doesn't support cohorts. The MCP no-start issue (acc-18) was closed as resolved internally, not confirmed in docs. source |
| Countly E Countly Ltd | ab_experiments_create makes a draft with targeting and ab_experiments_start begins it, but remote_config_parameters_update and conditions_update replace the whole resource, so other fields must be resent. source |
| Flagsmith B Bullet Train Ltd | create_feature, create_segment_override or versioned feature states and update_feature_state cover flags, but tools differ by v2 versioning, update_experiment only edits name and hypothesis, change requests are Enterprise, and a partial PATCH of a multivariate option returned 500 (issue 8901). source |
| GrowthBook BB GrowthBook, Inc. | Feature revisions add, update and publish rules in drafts and experiments start through /start or a status update, but some experiment fields replace the whole stored object, approval rules can hold writes, and the open MCP issue 58 says a plain flag needed a data source. source |
| Harness Feature Management & Experimentation BB Harness Inc. | fme_feature_flag update sends a merge-patch, definitions are created and updated, and segment keys are added by execute actions, but legacy and Harness-native scoping differ per resource, identity writes aren't implemented natively, and change requests can hold writes. source |
| Matomo B InnoCraft Limited | The AbTesting API manages experiments, but Matomo has no feature flags or percentage rollouts by targeting rule, and the create and start parameters couldn't be read. source |
| Mixpanel B Mixpanel, Inc. | Create- and Update-Feature-Flag and Update-Experiment (which manages lifecycle) exist, but audience and cohort targeting aren't supported through the MCP, and whether updates are partial isn't stated. source |
| Mostly Good Metrics E Mostly Good Metrics | create_experiment, update_experiment, start_experiment and stop_experiment cover the lifecycle, but there are no feature flags, targeting rules or rollout percentages, only variants. source |
| Optimizely Experimentation B Optimizely North America Inc. | Ruleset updates take JSON Patch and an enable endpoint turns a ruleset on, but the MCP's exp_manage_entity_lifecycle asks the person to confirm every live change, so unattended runs need the REST API. source |
| PostHog B PostHog Inc. (formerly Hiberly Inc.) | create-feature-flag, update-feature-flag, release-condition rollout and experiment-launch cover it, but feature-flag-rules-update replaces all rules, and the schema rejected a multi-value equals the UI accepts (acc-06, open). source |
| Statsig BB Amplitude, Inc. | Gates and experiments can be created and started (PUT /experiments/{id}/start), but gate_update and experiment_update replace the whole resource, partial edits need discover_tools, and updates require a confirmation. source |
Doesn't do this job 7
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| Fullstory B Fullstory, Inc. | No feature flags or experiments; the MCP only creates drafts of segments, metrics, funnels, journeys and dashboards. source |
| Heap D Content Square, Inc. | No feature flags or experiments. source |
| OpenPanel B OpenPanel AB | No feature flags or experiments. source |
| Pendo E Pendo.io, Inc. | No feature flags, and guide experiments are listed, not created or started, over MCP. source |
| RudderStack MCP C RudderStack, Inc. | No feature flags or experiments. source |
| Userpilot B Userpilot, Inc. | No feature flags or experiments in the MCP or API reference. source |
| Woopra D Woopra, Inc. | No feature flags or experiments. source |
What stops agents finishing
- an update replaces the whole resource and resets targeting
- the MCP schema rejects targeting the UI accepts
- starting an experiment only possible in the UI
- flag creation demands an experiment data source
- writes held for human approval or change requests
Around it
Before: code.add-flag-check. After: product-analytics.experiment-readout.
Where this comes from
Weight 15 of 100 in Product analytics & experimentation: 8 of 55 account links, nearly all GitHub issues about writes that silently did the wrong thing
The same as data: /api/v1/use-cases/product-analytics.flag-experiment-change.json, or the MCP tool get_use_case with id product-analytics.flag-experiment-change. Other jobs in this category: /api/v1/niches?category=product-analytics.
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.