Kimi K3 + Cuiar
Sixty listings in, one clean CSV out.
- ~/.kimi-code/mcp.json
- /mcp-config login
- Files in one --folder
Kimi Code runs K3 with a context of one million tokens, so long pages and big tables fit. Cuiar collects the rows from marketplaces and supplier portals in your own signed-in workspaces.
The files land in one folder. Kimi cleans them, compares them with last week and draws the chart.
Setup
Add Cuiar to Kimi Code.
Kimi Code runs on your computer, next to the Cuiar app.
- 1
Install Kimi Code
Then start
kimiand run/login. Homebrew and npm also work.curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash - 2
Add the server
Or run
/mcp-configin kimi. An entry withurland notransportis streamable HTTP.~/.kimi-code/mcp.json { "mcpServers": { "cuiar": { "url": "https://mcp.cuiar.com/mcp" } } } - 3
Sign in
Run this in kimi and approve on cuiar.com.
/mcpshows the status./mcp-config login cuiar - 4
Ask
List my Cuiar workspaces.
Other ways to connect
An API key
export CUIAR_API_KEY=cuiar_...with a key that has themcpscope.~/.kimi-code/mcp.json { "mcpServers": { "cuiar": { "url": "https://mcp.cuiar.com/mcp", "bearerTokenEnvVar": "CUIAR_API_KEY" } } }Local stdio
Needs the Cuiar CLI, signed in.
~/.kimi-code/mcp.json { "mcpServers": { "cuiar": { "command": "cuiar", "args": ["mcp"] } } }kimi.com and Kimi Work
kimi.com has no custom MCP servers. Kimi Work's Plugin Builder makes a plugin from an MCP URL; its sign-in to Cuiar is not tested.
Route
Kimi reads, Cuiar collects.
Kimi's own MCP calls use no Cuiar credits, and its context holds the long page text. For bulk runs, the CLI with --folder puts every file in one place.
- Remote MCPmcp.cuiar.com/mcp
- Best fit
- OAuth with
/mcp-config login, orbearerTokenEnvVar. - Local MCPcuiar mcp
- Works
cuiar mcpwith the CLI signed in.- CLIcuiar exec
- Works
- The Bash tool runs
cuiar run --folder ./data. - Routineson a schedule
- Works
- Kimi Code's cron jobs end with the session and after 7 days. A Cuiar routine keeps a daily check going.
Example tasks
Data from signed-in pages
01
Collect listings
cuiar run --workspace shop-research --folder ~/data \ "Collect the first 60 listings for ceramic pour-over coffee set on Amazon.de, Etsy and Otto: title, price, rating, review count, seller and URL. Save pour-over.csv."Kimi then cleans the file and charts the price spread.
02
Compare supplier prices
“In my Purchasing workspace, sign in to our four supplier portals. Read the unit price, minimum order and lead time for each SKU in skus.csv. Show what changed since last week's file.”
Run it with
--permission read-only.03
Match two languages
“For our 12 best-selling products, open the English and the German product page. Compare price, specs, image count and shipping text, and list each mismatch with both URLs.”
04
Find figures for a deck
cuiar --odyssey --permission read-only "Find the 2023 to 2026 market size, growth and top-5 brand share for European specialty coffee equipment, each with a source link."Kimi turns the figures into a 6-slide outline. Odyssey is on Pro and Team.
05
Check listing health
cuiar routines create "Listing health" --schedule "0 6 * * *" --time-zone Asia/Shanghai \ "Check our 30 listings on Amazon, eBay and Zalando. Report any that are not live, out of stock or without the buy box."
Good to know
- K3 and K3.1
- Moonshot lists K3 as the previous generation and kimi-k3.1-pro as the current model. Kimi Code runs both.
- Headless
kimi -pruns with auto permissions. Project MCP servers stay off until you trust the folder withKIMI_CODE_TRUST_WORKSPACE=1.- The old CLI
kimi mcp addexists only in the archived Python kimi-cli. Kimi Code usesmcp.jsonor/mcp-config.
Templates to run from Kimi K3
All templatesEach template page shows the run_task or create_routine call to give Kimi K3.
Engineering
Check web pages for accessibility problems
WCAG 2.2 AA problems on each page, with the element, the rule and a fix, the blockers first.
You start it: When you start it- Any website
- Google Docs
EngineeringAsks first
Triage new errors in Sentry
New and returning errors since the last run, with users affected, the suspect commit and draft issues.
On a schedule: Every weekday at 9:00- Sentry
- Linear
Engineering
Find the cause of a jump in your cloud bill
This month's cost per service against last month, with the resources that grew the most.
You start it: When you start it- Any admin panel
- Google Sheets
Engineering
Find why a deploy failed
The error from the build log, the change that caused it and a suggested fix, sent when a deploy fails.
On an event: When a deploy fails in Vercel- Vercel
- GitHub
Engineering
Search for fixes to an error message
Known causes and fixes for your error from forums, issues and docs, ranked by how well they match.
You start it: When you start it- Stack Overflow
- GitHub
- Any forum
Data
Fill in each new row of a sheet
The missing fields from official sources, with a source link for each value.
On an event: When a row is added to the sheet- Google Sheets
More agents
All AI agentsClaude CodeRemote MCPTest real flows from the terminal
CodexRemote MCPconfig.toml, codex exec and nightly checks
CursorRemote MCPCheck localhost in a real tab
Grok BuildRemote MCPAccessibility and QA passes on staging
ChatGPTRemote MCPErrands from a chat, with your OK on each click
OpenAI dotsCLIAn always-on dot, through your own computer
Start on the Free plan.
The MCP server, the vault and unlimited routines are free. Steps that your own agent takes use no Cuiar credits.