inventer-dev/mcp-internet-speed-test

Two deterministic, rule-based analysis engines for agents: cloud billing cost anomaly detection (rolling-median + MAD, resistant to the masking effect) and auditable architecture risk scoring (single point of failure, coupling, scalability gap) with a rule-by-rule breakdown for every score. `uvx cloudsealed-mcp`

Run efficiency for Claude Code sessions: the share of paid tokens that became model output versus re-reading context already sent, by tokens and cost-weighted. Reads local transcripts, makes no network calls.

Agent tool spend tracker. Local JSONL storage, consent-gated, nothing leaves the file. Free MCP server. `npx scvd-tab`

Cross-agent failure intelligence. Before an agent retries a failed tool call, `check_tool_failure` reports whether other agents are hitting the same failure right now and which recovery actually worked, with a Wilson-score confidence. Privacy-by-schema: no prompts, keys, or payloads stored. Hosted remote server at `https://failecho.com/mcp`, or self-host.

Governed managed-endpoint fleet operations (thin-client/VDI) — login-storm and patch/config-drift analysis, inventory, and guardrailed remediation (13 tools) with unbypassable audit logging (MCP + CLI), budget/runaway guards, and undo/rollback.

Prices every Claude Code and Codex session already on disk and lets the agent ask what it's spending mid-session: cost by session, sub-agent and model, week-on-week comparison, and plain-English findings with applyable patches. No proxy, no API key, nothing leaves the machine. Install: `brew install --cask magna-nz/tap/tallybook`.

Iris scores every agent run for quality, safety, and cost — on your machine, with no SDK and no account. Most agent projects check quality by running a few remembered prompts and eyeballing the output. Iris replaces that with numbers you can audit: your agent's runs land in a SQLite database on your disk, 13 built-in rules score them deterministically — PII, prompt injection, hallucination markers, cost thresholds — free, with no LLM calls, and an optional LLM judge with a hard per-eval cost cap handles the semantic questions. Every rule is inspectable and editable, because a judge you can't audit is just vibes with a number on it. MIT licensed, no telemetry; your traces never leave your machine.

6 tools · deploy_rule, evaluate_output

Your AI agent doesn't know what's broken in production. This fixes that. Last9 MCP Server connects Claude, Cursor, Windsurf, and any other MCP-capable AI assistant directly to your production observability data — logs, metrics, traces, exceptions, database queries, alerts, and deployments. The agent stops guessing and starts reading the actual signal. No binary to install. No tokens to manage. One URL, OAuth in your browser, done. Find your org slug in your Last9 URL: app.last9.io/<org_slug>/... Type /mcp, select last9, authenticate. That's it.

6 tools · get_alerts, get_database_slow_queries