Crowkis MCP · Model Context Protocol
Give your AI tools a memory.
Keep your tokens.
Claude Code and every MCP-capable app can hold the Crowkis cache as a tool: check it before spending tokens, bank every answer they compute. Repeated work becomes a local sub-millisecond hit — and nothing leaves your machine.
0 config block
is the entire integration
0 tokens
spent on a cache hit
0%
local — nothing leaves your machine
Your assistant repeats itself. Constantly.
AI coding assistants and agents re-ask the same questions with industrial enthusiasm: the same doc lookups, the same code explanations, the same boilerplate reasoning — dozens of times a day, billed at full token price every time. With Crowkis behind MCP, the model gains two reflexes: check the cache first, and bank what you compute.
In plain words: your AI assistant gets a memory that lives on your machine. Questions it has already answered stop costing you money — and your whole team shares the same memory.
Doc & API lookups
The same 'how does X work?' recurs across your whole team, all day. First ask pays; every ask after is free.
Code explanation
Explanations of stable code are stable. Cached until the file changes, then invalidated.
Agent tool results
Deterministic tool calls — schema fetches, searches — are pure savings on replay.
Multi-agent fan-out
Five agents asking variants of one question become one model call and four sub-millisecond hits.
The two-minute setup
Three steps, no SDK, no rewrite.
Run Crowkis
One docker pull, one docker run. The same instance can serve your app traffic and your AI tools at once.
Register the server
One config block — or one CLI line — and your assistant holds the cache as a tool.
Watch tokens stop burning
The assistant checks the cache before calling the model and banks what it computes. The dashboard counts the savings.
$ claude mcp add crowkis -- crowkis mcpOr add the config block by hand — works in any MCP-capable app, with a Docker variant if Crowkis runs in a container:
{
"mcpServers": {
"crowkis": {
"command": "crowkis",
"args": ["mcp"]
}
}
}Agents can't poison it
Every MCP write walks the same five-stage trust pipeline as all other traffic, with the assistant tracked as a source in the trust ledger. An agent that stores garbage earns a higher bar automatically — one assistant's hallucination never becomes the team's shared belief.
same gates · same ledger · same receipts in the dashboard