Show HN: Context Mode Insight – observability layer for AI coding agents

Context Mode Insight — AI engineering signal at $20/seat

the first Solution from Context Mode Platform · for enterprise AI engineering

Context Mode Insight — measure the engineering signal your AI tools are actually shipping.

Role-aware observability for Claude Code, Cursor, Copilot, Codex, Gemini, and 9 more AI assistants. 222 patterns. 13 MCP tools. Privacy-first. Sits on top of the open-source plugin 250,000+ developers already trust.

OSS collects. Platform understands. $20 / seat / month.

Start for $20 / seat → Or run the OSS plugin locally

⚡ open-source plugin · ELv2 → source on GitHub

250K+

devs on OSS

222

patterns

13

MCP tools

14

AI adapters

7

personas

Engineers at these companies run the open-source plugin locally.

Insight is the opt-in org-layer extension — Pro tier $20 per seat per month, self-serve at platform.context-mode.com.

MicrosoftGoogleMetaByteDanceRed HatGitHubStripeDatadogIBMNVIDIASupabaseAmazonCanvaNotionSalesforceHasuraFramerCursorVercelAnthropic

Open-source plugin = local-only by design.

Context Mode operates at the MCP protocol layer — raw data (web pages, API responses, file analyses, log files) stays in a sandboxed subprocess on the developer machine and never enters the LLM context window. The SQLite knowledge base lives in the home directory and dies when the session ends.

Nothing leaves the machine. No telemetry, no cloud sync, no usage tracking, no account required. The 250,000+ developers running the plugin today have never sent a byte to us.

Insight = explicit opt-in. Generate a token in the dashboard, the same plugin starts forwarding structured events (tool name, file path, error count — never prompt content, never source code). Remove the token, forwarding stops.

Pricing

One tier. Twenty dollars per seat. Predictable.

No annual variant. No volume discount. No founder discount. No grandfathering.

Pro

$20 / seat / month

222 patterns evaluated at org scope

MCP scopes — self, team, org

All 14 adapters supported

90-day data retention

Org Projects, Members, Teams, Untracked Activity

Persona views — Owner, Manager, Member

Email support (24-hour SLA)

Public cloud at platform.context-mode.com

Start for $20 / seat →

Per-seat means per member of your org (active or invited). Billed monthly. No annual variant, no volume discount, no free tier, no trial. Cancel any time via the Stripe Customer Portal — no refunds for the current period.

§ 01 · the problem

You shipped Claude Code to 50 engineers. Now what?

The seat license is signed. The plugin is rolled out. Engineering is moving faster — or it looks like it is. Then the questions start arriving from three different rooms at once.

CTO · Board deck Friday

"What is the AI investment actually returning? I need a number."

Bill is $50K/mo. Productivity claims float. No structured observability — only the invoice. Cannot defend the line item without a metric.

EM · Monday 1:1s

"Ahmet has been stuck on auth.ts for three days. Who else is stuck?"

Blockers visible only when someone Slacks for help. The quiet stuck stays invisible. Sprint slips. 1:1 prep is guesswork.

CISO · Audit window

"What did the AI actually touch? Where are the secrets?"

Sessions happen on every laptop, behind every API key. Compliance demands an audit log. The AI is faster than the audit trail.

Burning roughly $60K / year across 50 engineers on stuck context cycles — error loops, rewrites, re-prompts — that no dashboard surfaces.

Start for $20 / seat →

§ 02 · the solution

Context Mode Insight — the insight layer on top of the open-source plugin.

The context-mode OSS plugin already runs locally for 250,000+ developers. Every Claude Code, Cursor, Copilot, Codex, Gemini, Kimi, Qwen, JetBrains, Kiro session emits structured events through its hook system. Context Mode Insight ingests those events server-side and runs 222 patterns over them.

The findings surface in three places — same data, three doors in. Web Dashboard for the human eye. REST API for CI/CD and compliance pulls. Remote MCP for the AI agent reasoning over your engineering operations live.

Nothing on the developer laptop changes. The hook fires fire-and-forget on every tool call. Same plugin. Same workflow. Three new doors.

All operations are role-narrowed at the schema layer. A Member token cannot pass scope: "org" — the enum literally lacks the value. Defense at the schema, not the controller.

The OSS plugin stays free forever — it is the collector. The Platform is where your team sees what's happening. Pro tier is a flat $20 per seat per month. No tiers, no trial, no surprise metering.

01

Plugin

Local hook system emits structured envelopes — tool calls, file paths, error counts, token usage, commit messages. No prompt content, no file content.

02

Ingest

Cloudflare Worker validates with Zod, writes typed columns to D1. Streaming word-count UPSERT for §11 Layer-3 topics. Edge-cached reads.

03

Engine

222 patterns evaluated in batch via D1 native batch(). Sub-second p95 for org-rollups. Patterns derive intent from event correlation, not text vocabulary.

04

Surface

Dashboard (web). REST (CI / Slack / compliance). MCP (agent). Same data, three role-narrowed schemas.

Start for $20 / seat →

§ 03 · agentic by design

remote MCP — first-class agent surface

When the agent asks, the dashboard already answered.

Six personas. Six Mondays. One MCP endpoint, dynamically narrowed by role at issue time. Your CTO closes a board deck. Your EM preps a 1:1. Your CISO opens an audit window. Each agent calls the same server — and reaches a different surface.

Thirteen first-class tools — not an API wrapper. Sub-second cached responses. Works in Claude Code, Cursor, Codex, and every MCP-capable agent. Schema enforcement is structural — a Member token's find_blockers literally cannot accept scope: "org".

CTO · board prep // Quarterly board deck — ROI slide You: "what is our AI tool investment returning this quarter?"

Claude → engagement_health( scope: "org", period: "90d" ) ↓ 47/47 active · 100% adoptionHealth 87/100 · quality B+Productive rate 71% · adoption gap 12% Claude: Defensible board-ready snapshot. Adoption gap suggests a follow-up in next month's planning.

Insight cited: exec_org_health_score, exec_quality_grade, productive_rate_org, adoption_org_gap

EM · Monday triage // 1:1 prep before standup You: "who's stuck on my teams this week?"

Claude → find_blockers( scope: "team", period: "7d" ) ↓ 3 unresolved blockers · backend team12 persistent errors (high severity)Rejection volume 8 (warn) Claude: Start standup with the 3 blockers. Pair-program the error spike.

Insight cited: team_unresolved_blockers, team_persistent_errors, team_rejection_volume

IC · growth check // self-review before perf cycle You: "how am I doing vs my team?"

Claude → how_am_i(period: "7d") ↓ 78% productive sessions this weekHot file: auth.ts (consider split / review)Rules correlation 0.62 — above team median Claude: Solid trajectory for perf cycle. Watch auth.ts concentration.

Insight cited: personal_productive_session_new, personal_hot_files_new, personal_rules_correlation

CISO · audit window // monthly SOC2 evidence pull You: "any high-severity AI risks this month?"

Claude → identify_risk( scope: "org", period: "30d", severity: "high" ) ↓ 2 secret-shaped paths (auto-redacted)1 destructive command (sandbox scope)1 member · 8x MCP call rate (flagged) Claude: All caught at ingest. SOC2 evidence trail ready for the auditor.

Insight cited: secret_leak_proxy, policy_violation_destructive_command, mcp_tool_abuse_high_rate

FinOps · cost lens // CFO Friday — AI spend review You: "where is our AI spend going?"

Claude → top_patterns( scope: "org", category: "cost", period: "30d" ) ↓ Token spend +22% WoW (Opus-dominant)Cost per commit $0.84Retry waste $1,840/mo (Frontend team) Claude: Spend trending up — pair-program through the Frontend retry pattern.

Insight cited: token_spend_org_high, cost_per_commit_org, tool_waste_high_retry_cost

DevOps · release sign-off // pre-tag quality gate You: "is backend's commit quality healthy this week?"

Claude → top_patterns( scope: "team", team_id: "backend", category: "workflow", period: "7d" ) ↓ 41% feat · 28% fix · 22% refactor · 9% choreCadence 178 commits this weekFix ratio 28% (healthy band) Claude: Backend release sign-off green. Cross-language classifier holding up.

Insight cited: commit_healthy_mix, commit_strong_cadence, commit_fix_heavy

— all 13 first-class tools, one endpoint —

how_am_i

Personal snapshot + peer percentile

how_is_team

Single-team productivity + risks

compare_teams

Cross-team leaderboard

find_blockers

Ranked who is stuck and why

identify_risk

Severity-sorted risk surface

engagement_health

Adoption + ROI estimate

top_patterns

Exploratory entry, by scope

what_changed

Period-over-period deltas

top_members

Performers + at-risk leaderboard

activity_breakdown

Tool · platform · category mix

activity_timeline

Hourly · daily session series

member_detail

Single-IC deep drill (scope-aware)

commit_quality_team

Verb-family + Conv-Commits classifier

Start for $20 / seat →

§ 04 · persona × insight

Six people. Six Mondays. One data layer.

The same insight engine answers radically different questions depending on who is asking. Each persona's tool schema is dynamically narrowed at issue time — role is not a UI filter, it is a structural constraint.

Sarah · CTO

owner · org

"What is my AI tool investment actually returning? Show me org-level metrics for the last 30 days — I have a board deck Friday."

engagement_health(scope: "org", period: "30d")

Org ROI snapshot: 47 / 47 active. Adoption 100%. Estimated savings 34 hours / week / engineer (Think-in-Code substitution rate, high confidence). compare_teams shows Backend leading Frontend by +18% productivity, −12% error rate. Quarterly cost per resolved blocker: $42.

→ Renews the seat license. Board deck has a number.

Org pays 47 × $20 = $940 / month. Renewal is a Stripe quantity update, not a procurement event.

Brad · EM

manager · 3 teams

"This week's stuck list. Whom should I unblock first?"

find_blockers(scope: "team", period: "7d", limit: 5)

Top: Alex M. — error_cycle_member_high in auth.ts, 12 retries, 0 commits, severity high. Two more: a frontend dev hit rejection_loop on a CSS refactor (5d); a backend dev is on blocked_on > 14d for a migration. Server narrows results by managed_teams — Brad sees Backend, Frontend, Design only.

→ 1:1 schedule rewritten. Pattern broken before sprint end.

Alex · IC

member · self

"How am I doing? Am I above or below team average — and what should I work on?"

how_am_i(period: "7d")

Personal snapshot: productivity 78, quality 71. Strengths: parallel_tool_use_high, think_in_code_adopter. Growth: read_before_edit_skipped — 6× this week. Peer benchmark (anonymised): 62nd percentile. Recommendation: "When editing a file you haven't opened this session, read it first — your error_cycle rate drops 40% in those sessions."

→ Workflow adjusted. No one else sees the score.

Carter · CISO

manager · security

"What did the AI actually touch? Any secret patterns? Anyone running destructive commands?"

identify_risk(scope: "org", period: "30d", severity: "high")

3 high findings: secret_leak_proxy — 2 events surfaced .env-shaped basenames in file_paths (auto-redacted). policy_violation_destructive_command — 1 rm -rf in a sandbox. mcp_tool_abuse_high_rate — 1 member 8× normal MCP call rate. Audit-grade evidence count per finding.

→ Tickets filed. Audit window closes clean.

Sophia · FinOps

finance · cost

"Where is the AI spend going? Which teams are cost-efficient, which are over-prompting?"

top_patterns(scope: "org", category: "cost", period: "30d")

Top cost signals: token_spend_org_high trending +22% WoW (Opus 4.7 dominant). cost_per_commit_org $0.84 — within target. tool_waste_high_retry_cost — Frontend re

[truncated for AI cost control]