IF Whisperer: Parchment play with optional AI transcript hints via MCP

Technical Development Tools
1

*I’ve built IF Whisperer, an experiment for playing classic parser IF in the browser with optional AI help:
*
https://whisperer.solutions.cz/

It is not a new game and not a walkthrough site. It uses the Parchment-style browser play model: pick a game from a small shelf or paste an IFDB storyfile URL, then play normally in the browser.

The twist is that, when you get stuck, your own AI assistant can connect through an MCP server, read your transcript/session context, and give a nudge based on where you are, what you’ve tried, and what you’re carrying. The goal is “get unstuck without spoilers,” not to solve the game for the player.

I’d especially appreciate feedback from people who use or work on Parchment, Lectrote, Frotz, Gargoyle, IFDB play-online flows, or similar tools:

  • Is transcript-aware hinting useful for new parser IF players?

  • Does the AI layer feel like it belongs outside the interpreter, or should this stay completely separate from IF tooling?

  • What privacy/consent expectations would IF players have for transcript sharing?

  • Are there games or parser edge cases that would make good tests?

2

Neither your summery in this thread, nor the summery or terms on the linked site, state which LLM provider the MCP server is connected to. And that information is important so the end-user knows where to go to determine that provider’s transcript retention & usage policies, because some of the subject matter contained within some IF projects may be considered illegal in some Legal Regions.

eg. some providers retain transcripts, which can later be used by law enforcement as potential evidence of a crime being committed.

1 Like
3

Hello. this looks interesting, but is there any alternative than to sign in with Google?

4

This looks interesting, and I got it set up with one of the listed games on your site.

When I try to paste in a (working) link to an externally-hosted game, I get:

proxy fetch https://our-lady-of-thorns.netlify.app/thorns.z5: 400 {"error":"proxy_target_rejected","reason":"host","message":"host is not on the proxy allowlist"}

BrowsePlayMy sessionsSign out

Is this a problem on my side or yours?

5
  1. Probably, yeah. That’s what Invisiclues and similar systems aim to do. Some IF have in-game hint systems that track what has been accomplished.
  2. I’m not sure what you mean by this question.
  3. Everything should be able to remain client-side, or at the very least, private. The mere fact that I have to sign up (through Google or otherwise) is deterring me. Many people keep things entirely local, downloading files to their computers and playing with configured interpreters. Transcripts can be recorded automatically, but they should be entirely in the player’s control.
  4. I wonder how it would handle something like Counterfeit Monkey or Galatea.
6

Hello, the transcript is retained in the whisperer.solutions.cz PostgreSQL DB run by myself. The user can pick any AI tool capable of using the MCP protocol, I have tested mainly Claude, but other are possible and documented on the site as well (Codex, Cursor, etc. etc.). The transcripts are kept in the whisperer DB from where the user can delete them on request. You can even delete the whole account if needed. There is no way for me to persuade the users that I will not look into the transcripts. I guess the transcripts should not be harmfull for anybody but I migth be wrong here.

7

Currently there is no other way to authenticate than Google. I wanted to make it simple and what a lot of folks are already using.

8

I have a allow list for ifdb hosted games when proxying the request for binaries. So the problem is on my side as I wanted to avoid people misusing my site to free proxy of a random content. I might try to untighten the proxy a bit but I am afraid to allow just any random site for downloading the binaries thriough my tiny site.

9

For everything to stay client side - it might be a way harder to achieve because the AI assistants need to reach out somewhere when they want to access the transcripts or other queries (the queries are documented on the site). Reaching out to browser is not always feasible. That would have to completely change the architecture of the thing.

10

I meant that the Parchment based play is completely separate from your assistant session - different browser tab or even different browser where you can ask for the hint.

11

As the individual’s transcripts are stored remotely it is possible you may be contacted by law enforcement to supply a copy, assuming the relevant legal documents have been obtained and presented.

Have you considered how you’re going to response to & fullfil that legal request?
And in a way that doesn’t expose the transcripts of individuals not mentioned in the legal documents?