Atlas / MCP servers / newrelic-experimental / preflight

preflightBLOCK

Name: preflight
Author: newrelic-experimental

mcp/newrelic-experimental/preflight

MCP server by newrelic-experimental · ai agents · claude · claude code · Audited 2026-10-08

Observability for AI coding assistants — cost tracking, efficiency scoring, and anti-pattern detection for Claude Code, Cursor, Copilot, and more. Local-first; New Relic optional.

Verdict

BLOCK

Grade

Trust score

48 /100

Exposed tools

55 48r · 5w · 2d

Transport

stdio

License

Apache-2.0

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

Preflight Observability for AI Coding Assistants

[](LICENSE) [](.nvmrc) [](#quick-start) [](#dashboards)

**Docs** • **What's New** • Examples • **Community** • Contributing

Why Your AI Tool Needs Observability

Your AI coding assistant makes hundreds of decisions every session — what to read, what to edit, when to run commands. But you can't see any of it. You know it was fast, but was it efficient? You got a PR merged, but how much did it cost? You fixed a bug, but did it get stuck in a loop first?

Preflight is observability for agentic coding — the actions, cost, and efficiency of your AI coding assistant as it works. See exactly what's happening, how much it costs, and where your AI is wasting time.

Local-first by design. Preflight runs entirely on your machine and sends your data nowhere by default. A live dashboard at localhost:7777 shows your sessions in real time, fully offline. Connect a New Relic account only when you want more — team rollups, alerting, and cross-session history. You choose: local-only, New Relic, or both.

Demo

See cost breakdown, efficiency scoring, anti-patterns, and live session tracking in action.

What You Get

Visibility

Every action captured — file reads, edits, commands, searche

Read from source at commit 18f01b0ede1aOBSERVED · 2026-10-08

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code (npm)

claude mcp add preflight --env NEW_RELIC_LICENSE_KEY=${NEW_RELIC_LICENSE_KEY} -- npx -y @newrelic/[email protected]

Exposed tools (55)

48 read · 5 write · 2 destructive. Blast radius: 2 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`deep-research`	read	Fan-out web research workflow
`live-demo`	read	x
`live-fix-demo`	read	x
`nr_observe_get_anti_patterns`	read	Get detected anti-patterns (thrashing, re-reading, stuck loops, blind editing, over-delegation) for the most recent task.
`nr_observe_get_api_failures`	read	Get API failure tracking: per-model reliability scorecards, tokens lost, cost impact, throttle alerts, and mean time to recovery. Failures are captured via Claude Code
`nr_observe_get_budget_status`	read	Get current AI spend vs. configured budget caps (session, daily, weekly). Returns remaining budget, % used, and any threshold alerts fired this session.
`nr_observe_get_claudemd_impact`	read	Get the impact report for the most recent change to the project
`nr_observe_get_collaboration_profile`	read	Get a developer
`nr_observe_get_compute_waste`	read	Get compute waste summary: total tokens wasted on retried tool calls and anti-pattern activity (stuck loops, redundant reads, thrashing). Includes per-pattern breakdown and a status assessment (clean / moderate / needs_attention).
`nr_observe_get_config`	read	Show the current server configuration (sensitive fields masked): mode, developer, account, region, storage path, dashboard URL, and config file location. Use to diagnose misconfiguration without exposing credentials.
`nr_observe_get_context_composition`	read	Get per-turn token breakdown by category (system prompt, conversation history, tool results, injected files). Shows context window fill percentage and dominance alerts. LIMITATION: system_prompt and injected_file_content are always 0 -- the model API
`nr_observe_get_context_efficiency`	read	Get context window efficiency metrics: unique vs. repeated file reads, repeated-read ratio, and top re-read files. A high ratio suggests the model is losing context.
`nr_observe_get_context_tracking`	read	Get context window tracking: per-turn token growth, category breakdown (System/Tools/User/Assistant), fill percentage, and per-tool output contribution. Shows which tools consume the most context.
`nr_observe_get_cost_breakdown`	read	Get a breakdown of session costs by task, model, and efficiency metrics like cost per line of code. Cost figures are Preflight
`nr_observe_get_cost_forecast`	read	Project AI spending forward based on current session rate. Returns forecast cost for end-of-day, end-of-week, and end-of-session (8h), with a confidence note.
`nr_observe_get_cost_per_outcome`	read	Get cost breakdown by outcome type (bug fix, feature, refactor, etc.) with waste ratio and ROI estimate.
`nr_observe_get_cost_per_tool`	read	Cost attribution per tool type — approximate, based on turn-level token correlation. Shows which tools cost the most and average cost per call. Also breaks Skill calls down per skill in costBySkill (calls, estimated cost and tokens, total duration).
`nr_observe_get_decision_tree`	read	Get decision branch analysis: reasoning and action extraction per turn for triggered branches (recovery, retry, delegation), with outcome tagging. Includes post-mortem of failure chains and longest failure streak. LIMITATION: the reasoning field is the model
`nr_observe_get_efficiency_score`	read	Get the AI coding efficiency score for the most recent task or the session-wide rolling average.
`nr_observe_get_git_efficiency`	destructive	Get Git workflow efficiency metrics: merge conflicts, aborted operations, force pushes, stale branch detection, and actionable suggestions to reduce Git friction.
`nr_observe_get_instruction_drift`	read	Get instruction/prompt drift analysis: tracks how the project
`nr_observe_get_latency_decomposition`	read	Get latency decomposition: how much time is spent in LLM API calls vs tool execution vs overhead, with p50/p95 percentiles for each component.
`nr_observe_get_latency_percentiles`	read	Get p50/p95/p99 latency percentiles for tool calls, broken down by tool type. Use to identify which tools are slowest in the current session.
`nr_observe_get_model_recommendation`	read	Get a data-driven model recommendation ranked by historical efficiency score, cost, and task success rate across past sessions, both overall and broken down by task outcome type (bug_fix, feature, refactor, investigation, configuration, documentation, failed_attempt).
`nr_observe_get_personal_insights`	read	Returns a narrative coaching report comparing this week
`nr_observe_get_platform_comparison`	read	Compare AI coding assistant platforms side-by-side on a given metric: efficiency, cost, task_success, tool_calls, or error_rate.
`nr_observe_get_prompt_cache_health`	read	Get prompt cache health: hit rate, savings, and a concrete recommendation for improving cache efficiency.
`nr_observe_get_quality_proxy`	write	Get quality proxy metrics: diff apply rate, test pass rate, backtrack count, self-correction count, and degradation detection over the session lifetime.
`nr_observe_get_recommendations`	read	Get personalized optimization recommendations covering cost, efficiency, prompt engineering, CLAUDE.md impact, and model selection.
`nr_observe_get_retry_alerts`	read	Get thrashing/retry detection alerts: repeated failures or highly similar inputs within a sliding window. Identifies when the agent is stuck in a loop.
`nr_observe_get_session_history`	read	Get a list of past sessions with summary metrics (efficiency, cost, tool calls, outcome).
`nr_observe_get_session_stats`	read	Get current session observability metrics: tool call counts, success rates, file access stats, and duration summaries.
`nr_observe_get_session_timeline`	read	Get an ordered list of recent tool calls with timestamps, names, durations, and success/failure status.
`nr_observe_get_task_completion_rate`	read	Get task lifecycle metrics: completed task count, average task duration, and average tool calls per task.
`nr_observe_get_team_summary`	read	Get aggregated AI coding cost and efficiency metrics for all developers in the configured team, queried via New Relic NRQL. Requires teamId to be set in config.
`nr_observe_get_tool_selection_score`	read	Get tool selection quality score: evaluates efficiency of tool usage by detecting redundant reads, repeated failures, and unused large outputs. Score 0-1 where 1 is perfect.
`nr_observe_get_trends`	read	Get trend data for a metric over time: weekly efficiency, cost, task success rate, or tool call counts.
`nr_observe_get_turn_analysis`	read	Conversation turn analysis — groups tool calls by AI response, shows parallelism and turn patterns.
`nr_observe_get_weekly_summary`	read	Get a weekly summary report with per-developer breakdown, cost, efficiency, and anti-pattern counts.
`nr_observe_get_workflow_trace`	read	Get the complete tool call trace for a task, including sequence, duration, and anti-pattern/efficiency analysis.
`nr_observe_health`	read	Check server health: version, uptime, session ID, and connection timestamp. Also reports whether telemetry event sends are succeeding. Use when the MCP connection feels stale or tools are behaving unexpectedly.
`nr_observe_install_hooks`	write	Install PreToolUse, PostToolUse, PermissionRequest, and PermissionDenied monitoring hooks into ~/.claude/settings.json.
`nr_observe_mark_task_boundary`	read	Explicitly mark the end of the current task (a discrete unit of work between user
`nr_observe_report_feedback`	read	Record user quality feedback for a task. Helps correlate efficiency metrics with perceived quality.
`nr_observe_report_session_end`	read	Report that the current AI coding session has ended. Call this at the end of a session
`nr_observe_report_session_start`	write	Report that a new AI coding session has begun. Call this at the start of a session
`nr_observe_report_tokens`	write	Report token usage for cost tracking. Call periodically to enable accurate cost metrics.
`nr_observe_report_tool_call`	read	Report a tool call event for observability. Use this to report non-MCP tool calls
`nr_observe_send_digest`	write	Generate the current weekly AI coding summary and POST it to the configured Slack webhook immediately.
`nr_observe_subscribe_digest`	read	Register a Slack webhook URL to receive weekly AI coding cost and efficiency summaries.
`nr_observe_unsubscribe_digest`	destructive	Remove the registered Slack webhook for weekly digests.
`sample`	read	short
`tool_a`	read	A
`tool_b`	read	B
`tool_c`	read	C

Trust audit

BLOCKgrade F · trust 48/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (4 observation(s))
Network: declared (5 observation(s))
Shell: declared (1 observation(s))
Dependencies: not all pinned
Secrets in source: found

Findings (25)

HIGHFilesystem / path · fs.credential_store · CWE-22, CWE-59

src/platforms/amazon-q-adapter.ts:101

'   (typically ~/.aws/amazonq/mcp.json or project-level .amazonq/mcp.json)',

Why it matters. touches a credential store

HIGHFilesystem / path · fs.credential_store · CWE-22, CWE-59

src/platforms/amazon-q-adapter.ts:115

'   file (~/.aws/amazonq/cli-agents/<agent-name>.json for a global',

Why it matters. touches a credential store

MEDIUMNetwork egress · net.raw_ip · CWE-200, CWE-319

.claude/skills/verify-preflight/SKILL.md:30

$PF up                           # prints: up pid=<pid> url=http://127.0.0.1:7791 run=<run dir>

MEDIUMNetwork egress · net.raw_ip · CWE-200, CWE-319

.claude/skills/verify-preflight/scripts/pf-verify:12

URL="http://127.0.0.1:$PORT"

MEDIUMHard-coded secrets · secret.anthropic · CWE-798, CWE-321

src/transport/nr-ingest.test.ts:2676

agent_description: 'Use api key sk-ant-api03-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA',

MEDIUMHard-coded secrets · secret.anthropic · CWE-798, CWE-321

src/transport/nr-ingest.test.ts:3220

const SECRET = 'sk-ant-api03-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA';

MEDIUMHard-coded secrets · secret.db_uri · CWE-798, CWE-321

src/config.test.ts:747

const input = 'db: mongodb+srv://admin:[email protected]/db';

MEDIUMHard-coded secrets · secret.db_uri · CWE-798, CWE-321

src/config.test.ts:753

const input = 'db: mysql://root:rootpass@localhost/db';

MEDIUMHard-coded secrets · secret.db_uri · CWE-798, CWE-321

src/server.integration.test.ts:281

const fullDbUrl = 'postgres://user:[email protected]:5432/mydb';

MEDIUMHard-coded secrets · secret.generic · CWE-798, CWE-321

src/hooks/collector-script.test.ts:1907

const token = 'ghs_16C7e42F292c6912E7710c838347Ae178B4a';

MEDIUMHard-coded secrets · secret.generic · CWE-798, CWE-321

src/server.integration.test.ts:208

{ token: 'ghp_abcd1234efgh5678ijkl9012mnop3456qrst', name: 'GitHub PAT' },

MEDIUMHard-coded secrets · secret.generic · CWE-798, CWE-321

src/server.integration.test.ts:209

{ token: 'sk_live_abcd1234567890abcdefghijkl', name: 'Stripe live key' },

MEDIUMHard-coded secrets · secret.generic · CWE-798, CWE-321

src/server.integration.test.ts:210

{ token: 'pypi-AgEIcHlwaS5vcmc123456789abcdefghijklmnopq', name: 'PyPI token' },

MEDIUMHard-coded secrets · secret.generic · CWE-798, CWE-321

src/server.integration.test.ts:211

{ token: 'hf_abcdefghijklmnopqrstuvwxyzABCDEFG', name: 'Hugging Face token' },