LLMExpert: cited Q&A over YouTube videos MCP server
io.llmexpert/transcriptionllmexpert.iostreamable-httpv0.3.0Scanned
Cited Q&A over 2026 conference talks and YouTube videos. $0.02/query, free daily tier, key or x402.
14 tools: 6 change things and 8 only read.
- Reads private data · 1
- Reads text others write · 14
- Can send data out · 5
- 1 finding
Tools listed without sign-in. Remote posture
With Anlyon
6 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 LLMExpert: cited Q&A over YouTube videos MCP server can do
The LLMExpert: cited Q&A over YouTube videos MCP server is published by llmexpert.io, a domain the MCP Registry verified. It has 14 tools. It can create or change data (6 tools: query_channel, add_source, quote, transcribe and 2 more), but cannot delete data. 8 tools only read.
1 tool reads private data (list_sources); 14 tools bring in text other people can write (query_channel, list_channels, list_sources and 11 more); 5 tools can send data out of your system (query_channel, add_source, quote and 2 more). On its own it has all three parts of a prompt-injection exfiltration path: private data, text others write, and a way out.
The fixed checks matched 1 finding (annotation-conflict). Each one comes with the text it matched.
Its remote endpoint lists its tools without sign-in, and its tools changed 2 times since Anlyon Radar first saw them.
Last scanned Nov 11, 2026, version 0.3.0.
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.
14 tools
Changes things 6Claude Code: ask
add_sourceAdd a source to a channel and begin background ingestion.Reads text others writeCan send data outNo findings
Why it is in this group
- The name "add_source" says "add": 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.
- outboundThe input takes a destination (input.channel_id): what the tool writes goes there.
Description, as the server wrote it Add a source to a channel and begin background ingestion. Maps to POST /v1/channels/{channelId}/sources. Only `type: video` is processed today; `document` and `website` are reserved and return a 501 error until released. Charged from the key owner's prepaid balance at the ingest rate: $0.01/min standard, $0.03/min whisper by default, 1-minute minimum per video. A short balance returns a 402 error. Requires the key's `write` scope.
Annotations the server declares
- title: Add a source to a channel
- readOnlyHint: false
- openWorldHint: true
add_sourcesCharged like `transcribe`.Reads text others writeCan send data outNo findings
Why it is in this group
- The name "add_sources" says "add": 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.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Description, as the server wrote it Charged like `transcribe`. Expand the given YouTube URLs, charge for the new work and start an ingest job in an existing collection. Maps to POST /v1/collections/{collectionId}/sources. $0.01/min standard, $0.03/min whisper by default, 1-minute minimum per video. Videos already in the collection are free. Pass a `quote_id` from `quote` to be charged exactly that price. Poll `get_status` every 10–30 s until `status` is completed, partial or failed; `completed` means every accepted video is transcribed and indexed, so the collection is queryable. API key required (the key's `write` scope). Prices are the service defaults; `quote` returns the exact price.
Annotations the server declares
- title: Add YouTube sources to a collection
- readOnlyHint: false
- openWorldHint: true
create_collectionFree.Reads text others writeNo findings
Why it is in this group
- The name "create_collection" says "create": 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 Free. Create an empty collection to add sources to later with `add_sources`. Maps to POST /v1/collections. Returns the collection and its `collection_token` (`ct_…`, shown once: store it; absent on an idempotent replay). API key required (the key's `write` scope). Prices are the service defaults; `quote` returns the exact price.
Annotations the server declares
- title: Create an empty collection
- readOnlyHint: false
- openWorldHint: true
query_channelRun a retrieval-augmented query against a channel's indexed knowledge base and return a generated answer with citations.Reads text others writeCan send data out1 finding
Why it is in this group
- The name "query_channel" says "query": a word for reading.
- The description's first verb is "run": a word for a change.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe input takes a destination (input.channel_id): what the tool writes goes there.
Findings
- Warn
The server marks "query_channel" read-only (readOnlyHint: true), and its own name, description or input give the class "write".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"openWorldHint":true}; The name "query_channel" says "query": a word for reading.
Description, as the server wrote it Run a retrieval-augmented query against a channel's indexed knowledge base and return a generated answer with citations. Maps to POST /v1/query. Requires the key's `query` scope and access to the channel. Charged: $0.02 flat per call by default. Prices are the service defaults; `quote` returns the exact price.
Annotations the server declares
- title: Query a channel
- readOnlyHint: true
- openWorldHint: true
quoteFree.Reads text others writeCan send data outNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Description, as the server wrote it Free. Expand YouTube URLs exactly as an ingest would and return the price, the per-video lines and the per-source rejections. Maps to POST /v1/quote. $0.01/min standard, $0.03/min whisper by default, 1-minute minimum per video. The `quote_id` is valid for 15 minutes; pass it to `transcribe` or `add_sources` to be charged exactly this price. Pass `collection_id` to price against an existing collection (videos already in it are free). At least one of url, urls, channel_url, channel_urls. Requires the key's `read` scope. Prices are the service defaults; `quote` returns the exact price.
Annotations the server declares
- title: Quote an ingest
- readOnlyHint: true
- openWorldHint: true
transcribeCharged.Reads text others writeCan send data outNo 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.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Description, as the server wrote it Charged. Create a collection and start an ingest job for the given YouTube URLs in one call. Maps to POST /v1/transcribe. $0.01/min standard, $0.03/min whisper by default, 1-minute minimum per video. Pass a `quote_id` from `quote` to be charged exactly that price. Returns `job_id`, `collection_id` and `collection_token` (`ct_…`, shown once: store it; it reads this collection without an API key). Poll `get_status` every 10–30 s until `status` is completed, partial or failed; `completed` means every accepted video is transcribed and indexed, so the collection is queryable. With an API key it needs the key's `write` scope. Without an API key this tool is payable in USDC via x402: the first call returns a payment-required result (`isError` with the payment requirements in `structuredContent`); an x402-aware MCP client signs and retries the same call with `_meta['x402/payment']`, and the receipt returns in the result's `_meta['x402/payment-response']`. Never pay again after a 409 (already presented) or 502 (settlement unknown) result. Keyless `transcribe` takes one channel-like source and at most 3 videos. Prices are the service defaults; `quote` returns the exact price.
Annotations the server declares
- title: Transcribe into a new collection
- readOnlyHint: false
- openWorldHint: true
Only reads 8Show 8 read-only tools
get_sourceFetch a single source and its processing status (type-specific provenance included).Reads text others writeNo findings
Why it is in this group
- The name "get_source" says "get": a word for reading.
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 Fetch a single source and its processing status (type-specific provenance included). Maps to GET /v1/sources/{sourceId}. Requires the key's `read` scope.
Annotations the server declares
- title: Get a source
- readOnlyHint: true
- openWorldHint: true
get_statusFree.Reads text others writeNo findings
Why it is in this group
- The name "get_status" says "get": a word for reading.
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 Free. Get an ingest job started by `transcribe` or `add_sources`. Maps to GET /v1/jobs/{jobId}. Poll every 10–30 s until `status` is completed, partial or failed (terminal). `completed` means every accepted video is transcribed and indexed, so the collection is queryable; `partial` means some videos failed and were credited back (`credited_usd`); `videos.completed` / `videos.failed` count progress. Works with an API key or the collection's `ct_` token as the bearer. Prices are the service defaults; `quote` returns the exact price.
Annotations the server declares
- title: Get an ingest job
- readOnlyHint: true
- openWorldHint: true
get_transcriptFree.Reads text others writeNo findings
Why it is in this group
- The name "get_transcript" says "get": a word for reading.
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 Free. Get the transcript of a video that is completed in one of your collections. Maps to GET /v1/videos/{youtubeId}/transcript. `collection_id` is required with an API key; with the collection's `ct_` token as the bearer it defaults to the token's collection. `format` is json (segments with timestamps, default) or text (one `[H:MM:SS] text` line per segment). 404 when the video is not completed in that collection, or the transcript predates the transcript cache. Prices are the service defaults; `quote` returns the exact price.
Annotations the server declares
- title: Get a video's transcript
- readOnlyHint: true
- openWorldHint: true
get_usageGet usage and spend in dollars for the calling key: the account balance (wallet_usd), this period's spend (consumed_usd), query count, and budget remaining.Reads text others writeNo findings
Why it is in this group
- The name "get_usage" says "get": a word for reading.
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 Get usage and spend in dollars for the calling key: the account balance (wallet_usd), this period's spend (consumed_usd), query count, and budget remaining. Maps to GET /v1/usage. Requires the key's `read` scope.
Annotations the server declares
- title: Get usage and spend
- readOnlyHint: true
- openWorldHint: true
list_channelsList the channels the calling key is scoped to.Reads text others writeNo findings
Why it is in this group
- The name "list_channels" says "list": a word for reading.
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 List the channels the calling key is scoped to. Maps to GET /v1/channels. Cursor-paginated. Requires the key's `read` scope.
Annotations the server declares
- title: List channels
- readOnlyHint: true
- openWorldHint: true
list_sourcesList the sources (video/document/website) in a channel, optionally filtered by type or status.Reads private dataReads text others writeNo findings
Why it is in this group
- The name "list_sources" says "list": a word for reading.
What it can reach
- private-dataThe description's first sentence says "list" and "document": it reads documents.
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- untrusted-inputThe description's first sentence says "list" and "website": it reads web pages, which anyone can write.
Description, as the server wrote it List the sources (video/document/website) in a channel, optionally filtered by type or status. Maps to GET /v1/channels/{channelId}/sources. Cursor-paginated. Requires the key's `read` scope.
Annotations the server declares
- title: List sources in a channel
- readOnlyHint: true
- openWorldHint: true
query_collectionCharged: $0.02 flat per call by default, no token multiplier.Reads text others writeNo findings
Why it is in this group
- The name "query_collection" says "query": a word for reading.
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 Charged: $0.02 flat per call by default, no token multiplier. Ask a question of a collection and get a generated answer with citations. Maps to POST /v1/collections/{collectionId}/query. Query a collection once `get_status` says completed. With an API key it needs the key's `query` scope. Without an API key this tool is payable in USDC via x402: the first call returns a payment-required result (`isError` with the payment requirements in `structuredContent`); an x402-aware MCP client signs and retries the same call with `_meta['x402/payment']`, and the receipt returns in the result's `_meta['x402/payment-response']`. Never pay again after a 409 (already presented) or 502 (settlement unknown) result. Keyless, only the address that paid for the collection may query it. Prices are the service defaults; `quote` returns the exact price.
Annotations the server declares
- title: Query a collection
- readOnlyHint: true
- openWorldHint: true
search_talksAsk a question across about 490 curated 2026 conference talks and get a generated answer whose citations give each talk's video URL (with t=) and start_seconds, plus its conference…Reads text others writeNo findings
Why it is in this group
- The name "search_talks" says "search": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- untrusted-inputThe description's first sentence says "url": it reads web pages, which anyone can write.
Description, as the server wrote it Ask a question across about 490 curated 2026 conference talks and get a generated answer whose citations give each talk's video URL (with t=) and start_seconds, plus its conference, year and speakers. Optional filters: conference (one of ai-engineer-worlds-fair, ai-engineer-europe, kubecon-eu, google-cloud-next, microsoft-build, pycon-us, google-io, code-with-claude, openai-devday), year, and speaker (part of a name, case-insensitive; speakers are known for about half the talks, so a speaker filter misses the rest). A filter that matches no talk answers `answer: null` and costs nothing. Maps to POST /v1/search/talks. Charged: $0.02 flat per call by default, or free within the daily free allowance. With an API key it needs the key's `query` scope. Without an API key this tool is payable in USDC via x402: the first call returns a payment-required result (`isError` with the payment requirements in `structuredContent`); an x402-aware MCP client signs and retries the same call with `_meta['x402/payment']`, and the receipt returns in the result's `_meta['x402/payment-response']`. Never pay again after a 409 (already presented) or 502 (settlement unknown) result. Prices are the service defaults; `quote` returns the exact price.
Annotations the server declares
- title: Search 2026 conference talks
- readOnlyHint: true
- openWorldHint: true
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.
70 paths in this server alone from private data, through text someone else wrote, to a way out. Denying 1 private-data tool breaks every one.
- Reads private data1
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write14
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out5
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
{
"permissions": {
"deny": [
"mcp__transcription__list_sources"
]
}
}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, 19:13 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
- 6
- 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 6 tools that change things (6 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://llmexpert.io/mcp: tools/list returned 14 tools to an anonymous client, including query_channel (write), add_source (write), quote (write), transcribe (write), create_collection (write) and 1 more
Findings
Matches of fixed checks, with the text they matched.
- Warnquery_channel
The server marks "query_channel" read-only (readOnlyHint: true), and its own name, description or input give the class "write".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"openWorldHint":true}; The name "query_channel" says "query": a word for reading.
History
Changes to its tool definitions and to its remote endpoint, newest first.
Tools1 added, 1 changed.+1 added~1 changedView diff
- Added:search_talksOnly reads
- Changed:add_sourceChanges thingsdescription
Tools1 changed.~1 changedView diff
- Changed:get_usageOnly readsdescription
Install
Pinned to version 0.3.0. Values in angle brackets are yours to fill in.
claude mcp add --transport http transcription https://llmexpert.io/mcpclaude mcp add --transport http transcription https://llmexpert.io/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__transcription__list_channels",
"mcp__transcription__list_sources",
"mcp__transcription__get_source",
"mcp__transcription__get_usage",
"mcp__transcription__get_status",
"mcp__transcription__get_transcript",
"mcp__transcription__query_collection",
"mcp__transcription__search_talks"
],
"ask": [
"mcp__transcription__query_channel",
"mcp__transcription__add_source",
"mcp__transcription__quote",
"mcp__transcription__transcribe",
"mcp__transcription__create_collection",
"mcp__transcription__add_sources"
],
"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 transcription. Review it before you use it.
Capability tags behind 14 rules
- mcp__transcription__query_channel untrusted-input, outbound
- mcp__transcription__list_channels untrusted-input
- mcp__transcription__list_sources private-data, untrusted-input
- mcp__transcription__get_source untrusted-input
- mcp__transcription__add_source untrusted-input, outbound
- mcp__transcription__get_usage untrusted-input
- mcp__transcription__quote untrusted-input, outbound
- mcp__transcription__transcribe untrusted-input, outbound
- mcp__transcription__create_collection untrusted-input
- mcp__transcription__add_sources untrusted-input, outbound
- mcp__transcription__get_status untrusted-input
- mcp__transcription__get_transcript untrusted-input
- mcp__transcription__query_collection untrusted-input
- mcp__transcription__search_talks 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.llmexpert/transcription/verify?hash=a3af0f0f40d19787cd4b93c546d8cdb83fab34c38d8063939e0c658779236ac8" | jq -e '.changed == false'Pinned hash a3af0f0f40d1, from Nov 11, 2026, 08:18 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/io.llmexpert/transcription/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/io.llmexpert/transcription)Source: MCP Registry. Synced Oct 29, 2026, 20:18 UTC.
LLMExpert: cited Q&A over YouTube videos MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the LLMExpert: cited Q&A over YouTube videos MCP server safe?
- No scan can promise that. What its definitions show: 14 tools, 6 of which can change, send or delete something, and 1 finding from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the LLMExpert: cited Q&A over YouTube videos MCP server delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the LLMExpert: cited Q&A over YouTube videos 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 LLMExpert: cited Q&A over YouTube videos MCP server have?
- 14 tools: 6 write, 8 read. query_channel, add_source, quote, transcribe, create_collection, add_sources, list_channels, list_sources and 6 more.
Does the LLMExpert: cited Q&A over YouTube videos MCP server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the LLMExpert: cited Q&A over YouTube videos MCP server send data to outside services?
- 5 tools can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): query_channel, add_source, quote, transcribe and add_sources.
Has the LLMExpert: cited Q&A over YouTube videos MCP server changed its tools?
- Yes, 2 times since Anlyon Radar started tracking it, most recently on Oct 29, 2026. Every change is on this page and in its Atom feed.