crixin · coder mcp · capability module

Give your AI a memory.

Coder MCP is one half of the Crixin toolkit. Every Claude Code, Codex CLI, and Cursor session you've ever had — ingested into local SQLite, surfaced through four MCP tools: search_sessions, list_recent_sessions, get_session, stats. Stop re-explaining your project to Claude for the 40th time. The other half (Voice MCP) gives your AI a phone.

npm i -g crixin && crixin install Install in 30 seconds See Wrapped

Three jobs. Local. Indexed. Yours.

Your ~/.claude/projects/ folder is keeping a log on you. Coder reads it, dumps it into a local SQLite, and gives you a dashboard you can read in 30 seconds.

01

Recall

"Where did I solve that auth refactor?" Sub-second deep search across every Claude, Codex, and Cursor session — with copy-paste resume commands the second you find it.

02

Reflect

How you actually use AI: peak hours, model mix, project distribution, real cost per output token. Numbers you didn't know existed, computed locally on raw BPE.

03

Reveal

An honest behavioral archetype based on your actual sessions — Cowboy, Architect, Debugger, Tinkerer, Prompter-First. Plus an annual Wrapped report when December comes.

Eleven archetypes.

Five for you, six for your AI. Deterministic — no LLM, no telemetry. The rationale always cites real numbers from your data, never vibes.

You

Cowboyshort, fast, ship
Architectlong, deep sessions
Debuggerstack traces +1
Tinkerersmall, frequent
Prompter-Firstheavy user-led

Your AI

Skipper"pre-existing"
Apologizer"I apologize"
Hedger"I think" / "might"
Reverser"actually wait"
Optimist"all set" / "done!"
Stack-Tracerdebug rate ≥10%

Two commands. That's the whole thing.

No setup wizard. No API keys. No telemetry that magically "helps us improve." The package on npm is the entire product.

# Install both MCPs (voice + coder) into every supported host.
$ npm i -g crixin && crixin install

# Ingest your AI coding history into local SQLite.
$ crixin coder ingest

# Open the dashboard.
$ crixin coder dashboard

# Or just ask your AI host: "use crixin-coder to find when we last touched auth.ts"

What you'll see, specifically.

Concrete numbers from your real session history — not telemetry, not someone else's dataset. One command, one self-contained HTML file. Below: an actual sample of what crixin coder wrapped --year 2026 writes to disk.

crixin coder wrapped --year 2026
# → ./crixin-coder-wrapped-2026.html (118 KB, self-contained, no JS deps)
crixin coder wrapped · 2026

Your year of AI pair programming, in receipts.

91 sessions across Claude Code, Codex CLI, and Cursor. 17,156 messages exchanged. 2,744 hours of pair programming. $27.69 of real Anthropic + OpenAI spend, computed locally from token counts (no telemetry, no inference). Below is what got written to ./crixin-coder-wrapped-2026.html.

Sessions
91
across 3 hosts
Messages
17,156
avg 188 / session
Hours spent
2,744
~7.5 / day avg
Est. spend
$27.69
real BPE counts
Archetype
Architect
long, deep sessions
Hour-of-day activity
12a123456789101112p1234567891011
Day-of-week
Mon11
Tue14
Wed19
Thu22
Fri10
Sat8
Sun7
Monthly distribution
JanFebMarAprMayJunJulAugSepOctNovDec
Top projects
crixin2,134 msgs · $9.81
almadinashipping1,802 msgs · $6.40
haacpcompliance1,447 msgs · $4.92
personaldoppler988 msgs · $2.71
doppler (game)821 msgs · $1.94
compliantiep612 msgs · $1.34
Source breakdown
Claude Code63%
Codex CLI24%
Cursor13%
Top models
claude-opus-4-7$18.40 · 47%
claude-sonnet-4-6$6.91 · 31%
gpt-5$2.04 · 14%
claude-haiku-4-5$0.34 · 8%
PII auto-stripped · generated locally · shareable as a screenshot

Coder is the "being known" half of the Crixin brand: the only product that reads what your coding AI already wrote about you and tells you who it thinks you are. Voice MCP is the other half — the primitive that lets your AI use the phone.