Think back to the last AI-assisted coding session you had. You asked Claude or Codex or Cursor to help with something. It worked. You shipped it.
Now answer this: what exactly did it say? What did you ask, what did it propose, what did you approve? Can you reconstruct the reasoning chain, or is it gone the moment you close the tab?
Every other tool in your stack has an audit trail. Your git commits are logged. Your deploys are timestamped. Your database queries are traced. Your API calls are observable.
Your AI coding sessions, where increasingly consequential decisions are made, produce no persistent record whatsoever.
That's the gap Chron was built to close. And based on what's happened in the past few months, a lot of developers had the same feeling.
~7,500 downloads
v0.1.52 current version
5 AI clients supported
What Chron actually is
Chron is an MCP server that sits between you and your AI tools and logs every exchange, timestamped, cryptographically signed, stored in a local SQLite database that lives on your machine. No cloud. No subscription. No data leaving your environment.
One install. Then every Claude Code session, every Cursor agent run, every Codex task, every Gemini CLI interaction is captured automatically. You don't change how you work.
$ npm install -g chron-mcp
then restart your AI client
That's it. From that point, chron history shows you everything:
chron history
$ chron history
2026-08-14 09:12 claude 14 msgs · 8min Refactor auth middleware
2026-08-13 22:47 codex 22 msgs · 31min Review vendor onboarding policy
2026-08-13 18:03 cursor 9 msgs · 12min Add rate limiting to API routes
2026-08-12 14:55 claude 41 msgs · 1h4m Chron release gate — v0.1.50
4 sessions · 127 messages · 2.1h tracked
What changed between 0 and ~7,500 installs
The early versions were minimal: log the session, store the messages, expose a CLI. The core thesis was right, but the friction was real, developers had to manually configure each AI client and write instruction files by hand.
The last three months were about eliminating that friction entirely. Here's what shipped:
// Zero-friction install
npm install -g chron-mcp now configures everything, MCP server entries for every AI client you have installed, global instruction files (~/CLAUDE.md, ~/.cursor/rules/chron.mdc, ~/AGENTS.md), and permission fixes. The only required manual step is restarting your AI client, so it picks up the new MCP server.
// Session boundary integrity
The most important fix wasn't a feature, it was correctness. Early versions had a bug where resume latest would attach new, unrelated work to an old session from hours ago, polluting the audit trail. The fix introduced a three-guard system: a 30-minute time window, title pattern matching for compaction recovery, and same-tool scoping. Your sessions are now distinct records, not one long accumulation.
// Doctor, self-diagnosing setup
chron doctor now catches real-world problems that were previously invisible. The most important: when you have Chron configured in both user scope (npx -y chron-mcp) and local scope (a dev path), Claude picks the local one, it fails, and sessions silently don't log. Doctor now parses ~/.claude.json directly, detects the conflict, and prints the exact fix:
chron doctor
$ chron doctor
! Claude Code chron configured in conflicting scopes
user scope: npx -y chron-mcp
local scope: node /Users/you/chron/dist/index.js ← remove this
Fix: claude mcp remove chron -s local
// Multi-client project bootstrap
When Chron's MCP server starts inside a real project directory, it now automatically writes AGENTS.md and .cursor/rules/chron.mdc into that project. Codex and Cursor pick these up on the next session start. No manual chron connect needed per project.
// STORAGE
Local SQLite
All data stays on your machine. No accounts, no cloud sync, no telemetry.
// INTEGRITY
Ed25519 signatures
Sessions are cryptographically signed. Tamper-evident by default.
// PATTERNS
Behavioral detection
SIEM-compatible patterns flag prompt injection, credential exposure, scope creep.
// RISK
Attention scoring
Sessions get a deterministic risk score based on content, tool calls, and patterns.
// EXPORT
SIEM integration
Forward events to Splunk, Microsoft Sentinel, or CrowdStrike LogScale.
// CLIENTS
All major AI tools
Claude Code, Cursor, Codex, Gemini CLI, Claude Desktop — one install covers all.
The thing about ~7,500 downloads
I expected the early adopters to be security-conscious teams or compliance-driven orgs. They're not, or at least, that's not the majority. The bulk of installs are individual developers who want to know what their AI tools did.
The instinct driving it isn't compliance. It's the same instinct that makes you run git log before a release or check your query logs when something's slow. You want to know. The AI is making real changes to real code, and the conversation that produced those changes is currently ephemeral.
Chron makes it permanent. Not for auditors, for you.
Chron records. It does not certify. The record is yours, and it stays on your machine.
What's next
The logging layer is stable. The next layer is enforcement, runtime ALLOW/DENY gates that sit on top of the Chron proof and let teams define what AI tools are and aren't allowed to do. That's being built as CLAIIM, a companion project. Chron is the "what happened" record; CLAIIM is "was it authorized."
For now: if you're using any AI coding tool and you don't have a session log, you're flying blind. The install takes 30 seconds.
$ npm install -g chron-mcp
v0.1.52 · Node 18+
Start logging your AI sessions
Local-first · No accounts · Works with Claude, Cursor, Codex, Gemini
GitHub: https://github.com/nivaya/chron
npm: chron-mcp
Top comments (0)