Privacy Policy — Transcriptor MCP (hosted service)
Version: 1.4 Last updated: November 13, 2026 Effective date: December 14, 2026
Version 1.2 applies from December 4, 2026 until this version takes effect. If you buy the paid plan before December 14, 2026, the parts of this version about payment apply to you from the moment you buy.
Companion document: Terms of Service. "Output" has the meaning given to it in section 3 of the Terms.
1. Who is responsible
A natural person resident in Spain operates the hosted Transcriptor MCP service at https://transcriptor.gateway.mcpal.io/mcp (the "Service") and is the controller for the processing described here. Our postal address and tax identification number (NIF) are available on request. Email: [email protected]. We have not appointed a data protection officer and are not required to.
This policy covers only the hosted Service. If you run the software yourself from https://github.com/samson-art/transcriptor-mcp, we are not the controller and we process nothing about you.
2. Our role and MCPal's role
We publish the Service on the MCPal platform (mcpal.io), which we do not operate. MCPal supplies the gateway that relays your calls, sign-in through its Zitadel identity system, metering and quota enforcement.
MCPal is the controller for the account and identity data in its systems, for the acceptance records and for the per-call usage records in section 3: it shows both documents at sign-in, records your acceptance, and holds and deletes the usage ledger on its own systems and to its own retention rules. Its policy at https://mcpal.io/privacy.html covers that layer. We receive those records as a recipient, through the MCPal dashboard, and we are the controller for what our own server does with your request. If you send us a request we cannot answer because the data sits in MCPal's systems, we forward it and tell you we have done so.
The free allowance needs no payment, so no payment or billing data about you is processed for it. If you buy the paid plan, the operator of the MCPal platform sells it, and Stripe processes the payment. They are the controllers for the payment and billing data: your card details, your billing details and your receipts. We never receive your card details. From MCPal we receive at most the status and the dates of your subscription.
3. What is processed
- Account and identity data, held by MCPal: internal and Zitadel identifiers, email address, name, organisation identifier, creation and update dates. We never receive your password. The OAuth clients your MCP application registers automatically are stored with your identifier.
- Acceptance record (the platform calls it a consent record), held by MCPal: your Zitadel identifier, the server, the version of the Terms and of this policy you accepted, the time, your IP address and your browser user-agent string. It exists because both documents are shown on the sign-in screen, where you accept them.
- Per-call usage records: one row per tool call in MCPal's ledger, with the server, the tool name, the time, success or failure, the duration, a client reference, your identifier, and technical fields about the call that the gateway sets. These records identify you.
- Subscription and payment data, only if you buy the paid plan, held by MCPal and Stripe: your identifier, your email address, the identifiers of the Stripe customer and of the subscription, its status, the end of the paid period and whether you cancelled. Stripe alone holds your card details.
- What you send and what we return: the addresses, search terms and other parameters you submit, and the Output we return.
- Your MCPal identifier on our server: the gateway sends it to our server with each tool call. Our server does not store it and sends it to no one. Our server uses it only to make a pseudonymous identifier for the usage logs in section 6. A pseudonymous identifier is a code that only we can link back to you.
- Diagnostic data: the fault reports, performance traces and usage logs that our server sends to Sentry. Section 6 says what they hold.
4. What happens to your requests
Every tool call travels through the MCPal gateway to our server, so what you submit and the Output pass through both systems. Our server writes temporary files while it works and deletes them as soon as the call ends. If the server stops in the middle of a call, a file can stay until the server restarts and its temporary storage is replaced. We keep no Output beyond the cache described next. Both systems produce ordinary operational logs, so we do not promise that nothing about a request is written down.
The hosted Service keeps a response cache so that a repeated request is answered without asking the source platform again. A cache entry holds Output — caption text, a transcript, video metadata, the list of available subtitles or a chapter list — under the video address and the request parameters. It carries no identifier of the person whose request created it, and the same entry is served to anyone who makes the same request. Entries expire on their own after no more than 30 days. Search results and frames are not cached.
A run of frames from one video needs the same stream lookup each time, so our server keeps that lookup in memory to ask the source platform less often. An entry holds the video address you submitted, the requested width, the video identifier, its length, technical facts about the chosen format and the temporary stream addresses from the platform. It holds no frame and no identifier of the person whose request created it, and it serves anyone who asks for a frame of the same video at the same width. The server uses an entry for no more than 5 minutes. It removes the entry at the next frame request that needs a new lookup after that time, and at the latest when the server restarts. The entry does not go to the response cache or to disk.
If a video has no captions that we can use, our server can make a transcript of its speech. It downloads the audio of the video from the source platform to a temporary file and gives that file to a speech-recognition program (Whisper). This program runs on our own server, so no transcription provider receives the audio. It makes text from the speech and does not try to identify who speaks.
We transcribe only videos of up to 2 minutes. If the platform does not report the length of a video, our server downloads the audio, measures it and discards it when the video is longer. Our server never gives you the audio. It deletes the audio file when the transcription ends, and this can be after your call has ended. If the server stops before that, the file can stay until the server restarts and its temporary storage is replaced.
A transcript is Output. It goes to the user who asked for it and into the response cache described above, also when the transcription ends after the call. Other users who make the same request can then receive it.
We use nothing you send or receive to train any model, and we allow no provider to do so. We do no advertising, profiling or marketing.
5. Purposes and legal bases
- Give you access, run your calls, apply the daily free quota and apply a paid plan: performance of our contract, Article 6(1)(b) GDPR.
- Record the version you accepted, with IP address and user-agent string: our legitimate interest in proving the documents were shown and accepted, Article 6(1)(f). This is your acceptance of the documents, not consent under Article 6(1)(a).
- Keep the Service working, diagnose faults, prevent abuse and quota circumvention: our legitimate interest in a secure and reliable service, Article 6(1)(f).
- Count how many users meet errors, from the usage logs in section 6: our legitimate interest in a reliable service, Article 6(1)(f).
You may object to legitimate-interest processing at any time under Article 21 GDPR. We rely on consent for nothing described here, so there is no consent to withdraw. Our MCP endpoint accepts POST requests only and sets no cookies; any cookies at sign-in belong to MCPal.
6. Who receives data
MCPal (gateway, identity, metering, sale of the paid plan), Stripe (payment, only if you buy the paid plan), Zitadel (identity, used by MCPal), Supabase (the database holding the platform data in section 3), and the companies that host the MCPal gateway and our server. The hosting companies process every request and every Output in transit.
The Output goes back to the MCP client you connected, so the company that operates that client and its AI model receives it. That company is a separate controller under its own policy, we do not choose it, and we cannot control what it keeps. Consider this before you request material about identifiable people.
Our server sends diagnostic data to Sentry (Functional Software, Inc.), a service that collects errors and performance data. It sends three kinds of data:
- Fault reports: our server sends a report when a call fails because of a fault in our server or a failure of the source platform. A report holds the error, the address of our endpoint with any query string, and up to 100 log lines before the error. These are the lines of that request and lines that our server writes outside any request, such as its own checks. They can include the tool name and the video address and search terms that you submitted. They can also include the command that we ran and the error output of the source platform.
- Performance traces: for a sample of requests, our server sends a trace of the request, whether the call succeeds or fails. A trace holds the duration of the request and of its steps. It holds the same kind of log lines as a fault report, and the commands that our server sends to its response cache. A cache command holds the video address and the request parameters, not the Output. A trace names the version of the MCP protocol and a type of MCP client from a fixed list, such as "claude".
- Usage logs: our server sends one log for each tool call of a signed-in user, to count how many users meet errors. The log holds a pseudonymous identifier, the tool name, success or failure and the class of a failure. It also holds the kind of answer for a video without subtitles, the source platform, and whether a widget or the model made the call. The identifier is a keyed hash of your MCPal identifier, so Sentry cannot link it to you. A usage log holds no address, video identifier, search term or Output. Sentry does not link it to the fault reports and traces of the same call.
Fault reports and traces do not hold your name, email address or account identifier. They hold the identifier that the gateway gives each call, which the usage records in section 3 hold as the client reference. Those records link it to your account for 45 days, so we treat all three kinds as pseudonymous personal data. We send Sentry no request or response bodies, no cookies and no IP address from your request. From the request headers, Sentry gets only technical facts that do not name you, such as the call identifier and our host name. If the gateway or your MCP client sends Sentry trace headers, our trace joins the trace of the sender.
Sentry, Stripe and a hosting company may process data outside the European Economic Area, including in the United States. Where they do, the transfer is covered by the European Commission's standard contractual clauses or, where the recipient is certified, by the EU–US Data Privacy Framework. Ask us for the safeguards that apply to a given recipient and for the country its servers are in.
Our server requests captions, metadata, frames and audio from the source platform over our own network connection. We send that platform no identifier of yours and never use your credentials. We do not sell personal data, do not share it for advertising, and disclose it to an authority only where the law requires it.
7. How long data is kept
- Output: not kept beyond the call, except in the response cache. Temporary files are deleted at the end of the call, and at the latest when the server restarts.
- Response cache: no more than 30 days per entry, then expired automatically. Entries hold Output under the video address and request parameters, with no identifier of any user.
- Frame lookups: in memory only, used for no more than 5 minutes, removed at the next frame request that needs a new lookup after that time, lost on restart.
- Audio for transcription: a temporary file, deleted when the transcription ends, and at the latest when the server restarts.
- Usage records that identify you: 45 days, then deleted automatically. Before deletion they are rolled up into daily totals per server and per tool and into counts of distinct users per day, week and month. Those totals carry no identifier and are kept for the life of the Service.
- Subscription and payment records: kept by MCPal and by Stripe for the periods in their policies, which include the periods that tax and accounting law requires.
- Account and identity records: kept by MCPal while your account exists, for the periods in its policy.
- Acceptance records, including the IP address and user-agent string captured at sign-in: held by MCPal while your account exists and deleted with it, for the periods in its policy.
- Diagnostic data at Sentry: no more than 90 days for fault reports and performance traces, and no more than 30 days for usage logs.
- Operational logs on our server: no more than 30 days, then overwritten. MCPal's gateway logs follow the periods in its policy.
- Backups: overwritten on a rolling cycle of no more than 35 days.
8. Your rights
You may ask for access to all personal data we hold about you, and for correction, erasure or restriction. You may also ask for a portable copy of the data you gave us, and you may object to legitimate-interest processing. Write to [email protected]. We answer within one month, and may extend by two months for a complex request if we tell you inside the first month. We may ask for information to confirm the request is yours, because a user identifier is often the only information linking data to you. Identity and acceptance records sit in systems MCPal operates: we act on what we control and forward the rest.
You may complain to the Spanish supervisory authority, the Agencia Española de Protección de Datos (AEPD), C/ Jorge Juan 6, 28001 Madrid, www.aepd.es, which supervises us because we are established in Spain. You may also complain to the authority where you live or work.
9. Providing data, and automated decisions
An MCPal account and sign-in are needed to use the Service; without them we cannot provide it. Quota, paid plan, entitlement and abuse checks run automatically and decide only whether one request is allowed. A decision to suspend or end your access is taken by a person, not automatically. We take no automated decision producing a legal effect for you or similarly significantly affecting you.
10. People who appear in videos
Subtitles, transcripts, audio, metadata and frames can contain personal data about people who speak or appear in a video. We process it on our legitimate interest in producing the result the user asked for, Article 6(1)(f) GDPR. The data comes from the video at the address a user submits, on the platform hosting it, which our server reads at the moment of the request. It can include a voice in the audio and in its transcript, a name and other details in the description or metadata, and an image in a single still frame. It goes to the user who asked and to that user's MCP client. We keep it only as section 7 says.
Where a video published by the person concerned contains special category data, we rely on Article 9(2)(e) GDPR, data manifestly made public by the data subject. Where such data concerns anyone other than the person who published the video, we have no condition under Article 9 to process it, we do not knowingly process it, and section 8 of the Terms of Service forbids users to use the Service for that purpose. If you tell us that an address exposes special category data about you, we block that address.
We hold no contact details for the people who appear in videos, so we cannot notify them individually. Article 14(5)(b) GDPR allows this where notice would take disproportionate effort, so we publish this policy instead, at the address in section 12. If you appear in a video, you have the rights in section 8, you may object by writing to [email protected], and you may complain to the AEPD.
11. Children and security
The Service is not directed at children and we do not knowingly process their data; the minimum age is in the Terms of Service.
Traffic is encrypted in transit and access to our systems is restricted. We hold no security certification, and no online service can be guaranteed secure. Where a personal data breach occurs we notify the AEPD within 72 hours where Article 33 GDPR requires it, and tell affected users without undue delay where the risk to them is high. Report a security problem to [email protected].
12. Changes and contact
This policy carries a version number, and the acceptance record stores the version you accepted. We give at least 30 days' notice of any change, by email and additionally on the sign-in screen. If you do not accept the new version you may stop using the Service, which costs you nothing, and cancel a paid plan; section 18 of the Terms of Service applies to changes.
The current version is published at https://github.com/samson-art/transcriptor-mcp/blob/main/legal/PRIVACY_POLICY.md. Every version stays in that repository's history. This policy and the Terms are published in English.
Controller: the Service operator, Spain. Email: [email protected]. If a message to [email protected] comes back undelivered, ask us for our postal address and write to us there.