Context is this turn.
hexum_shrink returns what the model should see. Token counts come back on the call. There is no CLI and no browser IDE.
Not the repository. Hexum shrinks context on the path and fails a broken import before the model spends the tokens.
Agent dump
18.2k
Hexum MCP
3.1k
hexum_shrink returns what the model should see. Token counts come back on the call. There is no CLI and no browser IDE.
Layers live in architecture.hexum. The check is deterministic. It does not call a model. Drift fails the turn.
Mint hexum_live_ at /app. Paste once. The agent discovers tools from tools/list.
{
"mcpServers": {
"hexum": {
"serverUrl": "https://hexum.dev/mcp",
"headers": { "Authorization": "Bearer hexum_live_…" }
}
}
}Last updated 29 August 2026
What is Hexum How to cut coding-agent tokens MBPP and HumanEval numbers How those numbers were measured Pay for tokens you didn’t send FAQ
Hexum is an MCP server for Cursor, Claude Code, and other coding agents. It sits between the agent and the model: hexum_shrink sends this turn instead of the repository, hexum_check fails the turn when architecture.hexum is broken, and you keep your editor. On MBPP, context dropped about 85% with the same tests passing.
No. Hexum is a hosted MCP server. You keep Cursor, Claude Code, Antigravity, Codex, or Grok. Hexum is not an IDE and not a model.
The agent calls hexum_shrink before it reads widely. The model is sent this turn, not the repository. On MBPP, measured context dropped about 85% with the same tests passing. Example turn: 18.2k → 3.1k tokens, 41 files → 6. See /benchmarks and /methodology.
No. hexum_check is deterministic. It maps files to layers in architecture.hexum and reports forbidden imports. It does not call a model.
Hexum bills tokens saved: input minus what still goes to the model. Cache hits are free. No monthly cap. No seat tax. Usage early access from €0.30 per million. Teams from €0.35 per million. See /pricing.
Tokens in minus tokens out, when that is positive. Stripe is billed on tokens saved. Calls that do not save tokens still increment calls and the in/out counters.
No. Zero Data Retention is on Usage and Teams. MCP payloads are not stored. Hexum does not train on your code. Policy: /zdr.
Turn on usage billing at /app, mint a hexum_live_ key, paste the mcpServers JSON into Cursor Settings → MCP or ~/.cursor/mcp.json, then reload. Details: /cursor.
A file in the repo. Layers group files. Rules say a layer cannot use another. hexum_check reads it. Also accepted: hexum.hexum, .hexum, or the first *.hexum in the root. Guide: /architecture-hexum.
AUTH means the key is missing, invalid, or revoked. BILLING (HTTP 403) means that key’s billing source is off: personal Usage for a personal key, or Teams billing for a team key. Turn the matching plan on at /app, or mint a key against the plan that is on.
Keys from /app. Engine is MCP. Billing is usage, not a seat tax.
Get an API key