Amber

Anti-slop essay drafting and prose auditing. Grade A drafts, not Grade C.

Get your API key→

What it is

An editorial discipline shipped as an MCP connector for Claude. Plug it into any Claude environment that takes MCP — desktop, web, Code, Cowork — and you get two tools.

audit_amber — paste any prose, get a drift report. Em-dash density, throat-clearer count, bridging-tell scan, register fit. No LLM call. Free. No signup. Use it on your own writing, on AI output you're trying to clean up, on anything where you want to know whether the prose is disciplined or slop.

draft_amber — generate an essay from a brief, in one of six canonical essayist registers (Sontag, Baldwin, Didion, Camus, Le Guin, Woolf). Output is audited against the same anti-slop spec before delivery. Pricing scales with target length and mode: $1.50 for a 1500-word essay, $3 for 3000w, $6 for 6000w, $12 for a 12000w book chapter. Rewrite mode adds 50%; rewrite-with-continuity (cross-chapter awareness) adds another 20% on top, so a full 12000-word chapter rewrite with continuity is $21.60. $5 minimum credit purchase.

Get your API key

Who it's for

• Writers 30-40% through a manuscript who'd otherwise never finish
• Students with a sharp thesis and outline who need a workable first draft
• Anyone composing essays or book chapters up to ~12,000 words who'd rather hand-revise than hand-draft

Who it isn't for

• Full book-length manuscripts in a single call. Per-chapter drafting works (up to ~12000 words per draft_amber call, with continuity_context threading prior-chapter state through subsequent calls). Whole-book orchestration with subagent fan-out lives in the amber-voice skill.
• Voice replication. No per-user fingerprinting on this surface, and there won't be. Amber anchors output to literary masters, not to your own defaults.
• People who want a tool that confirms their existing prose. Amber refuses to amplify what's already there.

What it's honest about

• Output isn't voice replication. The host model can orient drafting toward your prior writing by feeding cues into the brief, but no per-user spec is stored on this server.
• The audit grades prose discipline, not argument. An Amber-perfect essay can still have a weak thesis. We grade slop, not reasoning.
• Above ~12,000 words per call the architecture stops fitting. Per-chapter drafting works (chunked-mode continuation tokens extend a single call up to that ceiling); use the skill for whole-book orchestration.
• The audit at v1 is literal-pass only — em-dash density, bridging-tell scan, throat-clearer scan. The paraphrase-class audit (semantic reword detection via a second model) is deferred.

The thesis stays yours. We never touch it.

Install

Tell us how it lands

Three questions. Replies open in your mail client and go to love@mahasangha.io.