LiveVariant MCP server
io.github.livevariant/livevariantlivevariantstreamable-httpv0.1.5Scanned
Build, inspect and read adaptive A/B tests. Config travels in the URL; no account or API key needed.
9 tools: 4 change things and 5 only read.
- Reads private data · 3
- Reads text others write · 5
- Can send data out · 0
- 2 findings
Tools listed without sign-in. Remote posture
With Anlyon
4 tools here change things. Govern them.
- Anlyon lists the tools and adopts none.
- Each tool you adopt gets approvals and receipts.
- Your agent never holds the key.
A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.
What the LiveVariant MCP server can do
The LiveVariant MCP server is published by the GitHub account livevariant. It has 9 tools. It can create or change data (4 tools: build_test, generate_priors, register_test and upload_image), but cannot delete data. 5 tools only read.
3 tools read private data (get_stats, get_test_status and variant_brief); 5 tools bring in text other people can write (get_stats, get_test_status, register_test and 2 more).
The fixed checks matched 2 findings (annotation-conflict and tool-shadowing). Each one comes with the text it matched.
Its remote endpoint lists its tools without sign-in.
Last scanned Nov 20, 2026, version 0.1.5.
Tools
Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.
9 tools
Changes things 4Claude Code: ask
build_testCreates a LiveVariant test and returns every URL needed to run it, plus a freshly generated stats secret.No findings
Why it is in this group
- The description's first verb is "create": a word for a change.
Description, as the server wrote it Creates a LiveVariant test and returns every URL needed to run it, plus a freshly generated stats secret. Pass `variants` to test one element, or `slots` to test several at once (hero image AND call-to-action, say). With slots the test optimizes the COMBINATION: one model learns how the elements interact, which two separate tests structurally cannot see. There is no algorithm to pick either way; every test runs the same joint model, sized from its shape. By default, nothing is registered anywhere: the config IS the test, encoded into the URLs, and the test's identity is a hash of it. Pass `publishableKey` on an account-enabled deployment to also register the new test to that key's organization, so it appears under My tests; the config and URLs are still the test, and registration failure is returned as a warning rather than failing the build. Editing a variant later produces a DIFFERENT test with its own empty history, which is usually what you want per campaign but is worth saying out loud to whoever you are building this for. The stats secret is returned once and never again. Only its hash goes into the config, so nobody, including this service, can recover it. Give it to the person who will read the results.
Annotations the server declares
- title: Build a test
- readOnlyHint: false
- destructiveHint: false
- idempotentHint: false
- openWorldHint: false
generate_priorsTakes YOUR estimate of how each variant will perform and converts it into the prior the model starts from, so a test does not spend its first visitors rediscovering what you alread…1 finding
Why it is in this group
- The name "generate_priors" says "generate": a word for a change.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
Findings
- Warn
The server marks "generate_priors" read-only (readOnlyHint: true), and its own name, description or input give the class "write".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":true,"openWorldHint":false}; The name "generate_priors" says "generate": a word for a change.
Description, as the server wrote it Takes YOUR estimate of how each variant will perform and converts it into the prior the model starts from, so a test does not spend its first visitors rediscovering what you already suspect. You supply the guess; this does the arithmetic and the capping. That capping is the point: a prior is expressed as pseudo-observations, and it is deliberately held weak enough that real data overrides it quickly. The response says exactly how many real visitors per variant it takes to wash your guess out, so you can judge whether you have been too confident. Being wrong here costs a little early traffic, not the test. Priors are outside the identity hash, so the test keeps its id, its URLs and any history it already has. Pass `when` to make the belief hold for ONE segment only ("image B is the one for the blue segment"). Without it the belief is about every visitor, which is a different and much stronger claim.
Annotations the server declares
- title: Add warm-start priors
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
register_testRegisters a test you built earlier to the organization a publishable key belongs to, so it shows under My tests and its stats are readable from the dashboard without the secret.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Registers a test you built earlier to the organization a publishable key belongs to, so it shows under My tests and its stats are readable from the dashboard without the secret. Use this only when the user provides both the test's stats secret and a publishable key for an organization they administer. The stats secret must match the hash inside the config, and the publishable key identifies the organization to register into. Prefer passing publishableKey to build_test directly: it registers at creation in one step. Keyless tests cannot be registered this way (nothing to prove with); they register through the tag on a verified domain. The organization can remove a listing from its dashboard (the test itself keeps serving).
Annotations the server declares
- title: Register a test to an account
- readOnlyHint: false
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
upload_imageUploads an image to the deployment's asset store and returns its URL, for use as a variant's `image` (email tests) or `url`.Reads text others writeNo findings
Why it is in this group
- The name "upload_image" says "upload": a word for a change.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Uploads an image to the deployment's asset store and returns its URL, for use as a variant's `image` (email tests) or `url`. The returned URL is deliberately not fetchable on its own: assets are only served with a short-lived signature that the serve endpoints mint per request, so uploading here does not create free static hosting. Use `previewUrl` (valid for an hour) to check what was stored. Storage is content-addressed: the id is the sha256 of the bytes, so uploading the same image twice is harmless and returns the same URL. Raster images only; SVG is refused because it can carry scripts. Not every deployment enables asset hosting, and this tool says so plainly when yours does not.
Annotations the server declares
- title: Upload an image
- readOnlyHint: false
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
Only reads 5Show 5 read-only tools
get_statsFetches a test's results and works out what they mean.Reads private dataReads text others writeNo findings
Why it is in this group
- The name "get_stats" says "get": a word for reading.
What it can reach
- private-dataThe input names a secret (input.statsSecret), and the tool reads: it reads secrets.
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Fetches a test's results and works out what they mean. Alongside the raw counts it returns the probability that each combination is genuinely best and the expected cost of stopping now and keeping the leader. Use those rather than comparing conversion rates by eye: a variant ahead 2/10 to 1/10 looks twice as good and is very close to a coin flip, and that mistake is the single most common way an A/B test gets called wrong. Multi-slot tests also report per-slot marginals: how each variant did across every combination it appeared in. Needs the stats secret. If you have the manage URL, its #fragment IS the secret and it will be used automatically.
Annotations the server declares
- title: Read a test's results
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: true
get_test_statusReports what the deployment's registry knows about a test: whether it is claimed into an account (and by which organization), and whether each redirect destination is a verified do…Reads private dataReads text others writeNo findings
Why it is in this group
- The name "get_test_status" says "get": a word for reading.
What it can reach
- private-dataThe input names a secret (input.statsSecret), and the tool reads: it reads secrets.
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Reports what the deployment's registry knows about a test: whether it is claimed into an account (and by which organization), and whether each redirect destination is a verified domain. Unverified destinations work, but visitors see a 'Redirecting you to…' continue screen first. When you see verified: false, tell the user to verify the domain under Settings on the dashboard; the three ways are a DNS TXT record, serving the well-known file, or having the SDK tag with their publishable key live in the site's source. If the test is unclaimed, remind them the manage URL claims it in one click when opened signed in. Requires the test's stats secret, the same as get_stats. A manage URL's #fragment is used automatically.
Annotations the server declares
- title: Check a test's account and domain status
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: true
inspect_testDecodes a test and describes it: slots, variants, context, and whether it can be served by redirect.No findings
Why it is in this group
- The name "inspect_test" says "inspect": a word for reading.
Description, as the server wrote it Decodes a test and describes it: slots, variants, context, and whether it can be served by redirect. Also lints it for the mistakes that only show up once a campaign is out, such as an email test whose context comes from geo (which a mail proxy answers about itself). Use this before sending anything, and to answer 'what is this link?'.
Annotations the server declares
- title: Inspect a test
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
list_testsLists tests registered to the signed-in account, newest first, with cursor pagination and an optional case-insensitive name filter.1 finding
Why it is in this group
- The name "list_tests" says "list": a word for reading.
Findings
- Warn
The description contains "every other tool", wording about other tools or servers.
What this check looks forMatched text
description: "...nswers for an identified caller: unlike every other tool, WHOSE tests these are cannot be expres..."
Description, as the server wrote it Lists tests registered to the signed-in account, newest first, with cursor pagination and an optional case-insensitive name filter. Only exists on deployments with accounts, and only answers for an identified caller: unlike every other tool, WHOSE tests these are cannot be expressed as an argument. Each entry carries the encoded config, which inspect_test and get_stats accept directly.
Annotations the server declares
- title: List my tests
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
variant_briefReturns the constraints to write or generate test variants against, for email or web, plus the rules that decide whether a test can be read at all once it runs.Reads private dataReads text others writeNo findings
Why it is in this group
- The description's first verb is "return": a word for reading.
What it can reach
- private-dataThe description's first sentence says "return" and "email": it reads mail.
- untrusted-inputThe description's first sentence says "return" and "email": it reads inbound email, which other people can write.
Description, as the server wrote it Returns the constraints to write or generate test variants against, for email or web, plus the rules that decide whether a test can be read at all once it runs. The one that matters most: one idea per slot. To vary two elements, give the test two slots and let it learn the combination, rather than bundling both changes into one variant and never learning which half worked. Ask for this before drafting variants, then produce them yourself against what it returns.
Annotations the server declares
- title: Brief for writing variants
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.
Exfil surface
Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.
No private-data, untrusted-input, outbound path in this server alone: no tool is tagged outbound.
- Reads private data3
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write5
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out0
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.
Remote posture
What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.
Looked at Nov 20, 2026, 19:07 UTC, unchanged since Oct 22, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- yes
- HTTP status of an anonymous initialize
- 200
- tools/list served without credentials
- yes
- Tools that change things, listed without credentials
- 4
- Protocol version
- 2025-06-18
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- not observed
- Authorization servers
- not observed
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- not observed
- Client ID metadata documents
- not observed
- WarnThe whole server
The endpoint served tools/list without credentials, and the list has 4 tools that change things (4 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://livevariant.com/mcp: tools/list returned 9 tools to an anonymous client, including build_test (write), generate_priors (write), register_test (write), upload_image (write)
Findings
Matches of fixed checks, with the text they matched.
- Warngenerate_priors
The server marks "generate_priors" read-only (readOnlyHint: true), and its own name, description or input give the class "write".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":true,"openWorldHint":false}; The name "generate_priors" says "generate": a word for a change.
- Warnlist_tests
The description contains "every other tool", wording about other tools or servers.
What this check looks forMatched text
description: "...nswers for an identified caller: unlike every other tool, WHOSE tests these are cannot be expres..."
History
Changes to its tool definitions and to its remote endpoint, newest first.
No change seen yet. Its tools are rescanned every day, and a remote endpoint is looked at once a week.
Install
Pinned to version 0.1.5. Values in angle brackets are yours to fill in.
claude mcp add --transport http livevariant https://livevariant.com/mcpclaude mcp add --transport http livevariant https://livevariant.com/mcpRun in a terminal.
Claude Code permissions
Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.
{
"permissions": {
"allow": [
"mcp__livevariant__inspect_test",
"mcp__livevariant__get_stats",
"mcp__livevariant__get_test_status",
"mcp__livevariant__list_tests",
"mcp__livevariant__variant_brief"
],
"ask": [
"mcp__livevariant__build_test",
"mcp__livevariant__generate_priors",
"mcp__livevariant__register_test",
"mcp__livevariant__upload_image"
],
"deny": []
}
}Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as livevariant. Review it before you use it.
Capability tags behind 5 rules
- mcp__livevariant__get_stats private-data, untrusted-input
- mcp__livevariant__get_test_status private-data, untrusted-input
- mcp__livevariant__register_test untrusted-input
- mcp__livevariant__upload_image untrusted-input
- mcp__livevariant__variant_brief private-data, untrusted-input
Pin
Keep the decision: fail CI when the tools change, follow every change, or show a badge.
Pin in CI
Fail a build when the tools you reviewed are not the tools the server serves today.
curl -s "https://radar.anlyon.com/api/v1/servers/io.github.livevariant/livevariant/verify?hash=46637a4e5ca91b91798865b9922e9a5e4f6523ef3f1b6b524ea19bb89357b6c7" | jq -e '.changed == false'Pinned hash 46637a4e5ca9, from Nov 20, 2026, 01:18 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/io.github.livevariant/livevariant/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/io.github.livevariant/livevariant)Source: MCP Registry · repository. Synced Oct 21, 2026, 18:38 UTC.
LiveVariant MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the LiveVariant MCP server safe?
- No scan can promise that. What its definitions show: 9 tools, 4 of which can change, send or delete something, and 2 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the LiveVariant MCP server delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the LiveVariant MCP server do things that cannot be undone?
- No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.
What tools does the LiveVariant MCP server have?
- 9 tools: 4 write, 5 read. build_test, generate_priors, register_test, upload_image, inspect_test, get_stats, get_test_status, list_tests and 1 more.
Does the LiveVariant MCP server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the LiveVariant MCP server send data to outside services?
- None of its tools sends, posts or publishes, or takes a recipient, channel or URL.
Has the LiveVariant MCP server changed its tools?
- Not since Anlyon Radar started tracking it. Last scanned Nov 20, 2026.