
Build rich applications with real-time social data. Powered by @LunarCrush.
California, United States
Joined July 2019
- Tweets13.1K
- Following4
- Followers15.7K
- Likes7.4K
A single alert threshold is really a weekend detector.
/topic/minecraft/time-series/v1, bucket day, 89 full days to Oct 23.
Mean engagements: Sunday 968M, Saturday 962M, Tuesday 817M. All days together, 865M.
Group by weekday, then alert.
The whole alert: daily buckets, grouped by weekday, yesterday against its own median.
lunarcrush.com/en/developers…
A weekend build with no JSON parsing in it.
lunarcrush.ai serves the same data as LLM-ready markdown, addressable by path. /list/categories came back this morning as a ranked table of all 41 categories.
Point the agent at the path. Skip the client library.
A sentiment percent with no mention count beside it is noise.
/topic/runway-ai/v1 returned 90% sentiment as of this morning, across 144 mentions from 39 accounts.
/topic/open-ai/v1 returned 77%, across 92,458 mentions.
Read sentiment and num_posts together.
"Which network is driving this?" should not cost a sprint.
/topic/tesla/v1 returns all six. As of this morning X held 69% of the 91,963 mentions and 21% of the 301.7M interactions. YouTube: 12% of mentions, 39% of interactions.
X-only monitoring sees the talk, not the reach.
What this replaces, and what it plugs into:
lunarcrush.com/en/integratio…
Ask Claude which gaming topics sit furthest below their category on sentiment.
The MCP list tool reads /category/gaming/topics/v1 and answers in one turn.
Gaming ran 69% sentiment as of this morning. Resident Evil ran 21% across 15,596 mentions.
One outlier, no dashboard.
Every spike alert on a daily series ships the same bug. The last row is a partial day, so today always looks like a collapse.
Drop it, take the median of what is left, fire at 1.2x.
On the anthropic topic: median 101.4M interactions over 29 complete days, 3 days above the line.
One command returns all six networks. No per-network auth.
lunarcrush topic porsche
As of this morning X carried 37% of mentions and 7.8% of interactions. TikTok carried 34% of mentions and 48% of interactions.
Where people talk is not where reach lands.
One line installs it, and it ships with every paid plan:
lunarcrush.com/en/developers…
Follower count is a weak filter for influence.
/topic/roblox/creators/v1 returns followers next to 24h interactions.
As of this morning the top creator held 842K followers and 93.2M interactions. A 15.1M-follower account in the same table: 5.8M.
Rank by delivered reach.
The endpoint page, with roblox already in the query string:
lunarcrush.com/en/developers…
One dataset in three places is usually three integrations.
The pipeline reads JSON. The analyst lives in a terminal. The assistant speaks MCP.
One LunarCrush key covers all three, same Bearer header. Five active keys, so each environment holds its own.
Keys are managed on the API Keys page, five active at a time.
lunarcrush.com/en/developers…
Four ways a correct API response still looks wrong.
/coins/list/v1 is cached up to an hour, v2 refreshes every few seconds. The last row of a daily time series is a partial day. And /system/changes logs edits that move historical numbers.
The generated field dates the payload.
Field list and the freshness note for the uncached version: lunarcrush.com/en/developers…
Rate limits decide your architecture before your code does.
Hobby: 4 requests a minute, market data only. Individual: 10. Builder: 100 a minute, 20,000 a day, every endpoint. Scale: 500 and 100,000.
A nightly batch over 500 topics fits Builder. A live per-user panel does not.
Three hours of glue code to give an assistant social data, or one config block.
The MCP server is a URL and a bearer header. Claude Desktop reaches it via mcp-remote. Cursor and any client that takes a URL points straight at it.
One key covers the API, the CLI and MCP.
The exact config for both clients, and where the key comes from: lunarcrush.com/en/developers…













