ListeningKitMemelord
Memelord (for developers) is a library of meme templates, video clips and stickers, searchable by meaning, with what each one means and what is trending now. log uses it to answer a chat with a meme that fits the moment (rule 119).
What log can do with it
Only Memelord's four free discovery tools are reachable from log, so nothing costs money (rule 41). Generating, editing or running any model is not wired in.
| Tool | Command | What it gives |
|---|---|---|
memelordSearch | log memelord search "<the moment>" [--type meme|video|sticker] [--method context|combined|media|fulltext|fuzzy] [--limit <n>] | Templates that fit, each with its name, type, media url and description |
memelordTemplate | log memelord template <id> | One template in full: its media url, text slots and description |
memelordTrending | log memelord trending [--limit <n>] [--media image|video] | What is trending now, each with its cultural context |
memelordTools | log memelord tools [--category imageToImage|imageTextToImage|textToImage] | The catalog of one-shot image tools; log lists them and never runs them |
Memelord's REST API (https://www.memelord.com/api/v1) is called with plain GETs, the key sent as a bearer token. Its own camelCase-breaking values (image-to-image, semantic_context) are converted at that boundary only.
matt's meme taste
A meme that fits the moment still has to be his kind of meme. matt named his kind (2026-11-02): deep fried memes, memes about larping, streamer reaction clips like FlightReacts, Zias and Bloo, and shitposts. log keeps that as his taste (rule 120):
- The index.
log memelord taste --build --from <posts.json>reads every picture in his X media posts (fetched with treg'stikhub.x.twitter-web-fetch-user-media, about $0.001 a page) and asks SubQ, from the picture itself, whether it is a meme and which of his styles it is. The memes are kept in~/.logCli/memes/taste.json(LOG_MEME_TASTEmoves it). The first build, 2026-11-02, read his 125 media pictures. - Ranked search.
log memelord searchasks SubQ, for each template it finds, whether the template's picture is his kind of meme, given his styles and the memes he has sent, and ranks by that score (taste 0.82 his kind, streamerReacts).--noTastekeeps Memelord's own order. Memelord's search knows nothing about these styles ("deep fried" finds Burnt Patrick), which is why the taste reads the picture. - In context. Every search answer leads with the taste line, and
log memelord tasteshows it with the memes he has sent, so whoever picks a meme has his taste in front of them.
Setting it up
- Create a key at memelord.com/settings/developer. It starts with
mlord_live_. - Put it in
.env.localasMEMELORD_API_KEY=…(or in~/.config/listeningkit/memelord-api-key.txt). - Restart the Claude app so the log MCP has the tools.
Without a key every tool answers memelordKeyMissing and calls nothing. Memelord's own errors are passed on with their code (API_KEY_INVALID, NOT_FOUND, INVALID_PARAMETER).
Sending a meme
- Discord: the meme goes out as its media url on a line of its own; Discord shows an image (
.webp,.png) or a clip (.mp4) inline. - X DMs:
twitterDmLivetakesmedia, a file or the meme's url. A url is downloaded first, uploaded to X as DM media (dm_image,dm_gif,dm_video) and attached to the first message of the dm, so it shows as a picture, never a link. First sent 2026-11-02 to @gabesweats: Waiting Skeleton, with "me waiting on your bug report / you break it yet lad". The whole loop, read, pick, search, gate, send, is the use case docs/useCases/discord/memeReply.md and thememelordskill.
How it is checked
The spec is docs/specs/memelord.md: the code proves the four endpoints and their GET method, and SubQ checks the behaviour on every commit (rule 117).