Skip to content

metrxbots/mcp-server

v0.2.5 Feature

This release adds 3 notable features for engineering teams evaluating rollout.

Published 5d MCP Developer Tools
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

ai-agent cost-optimization finops llm mcp mcp-server

Summary

AI summary

Updates Highlights, https://github.com/ckpark123/Metrx/pull/223, and correctness across a mixed release.

Changes in this release

Feature Medium

`get_attribution_report` now consumes flat `/api/outcomes` list and aggregates client‑side

`get_attribution_report` now consumes flat `/api/outcomes` list and aggregates client‑side

Source: llm_adapter@2026-05-29

Confidence: high

Feature Medium

`get_task_roi` supports both `days`‑only and explicit `start_date`/`end_date` date ranges for legacy and new backends

`get_task_roi` supports both `days`‑only and explicit `start_date`/`end_date` date ranges for legacy and new backends

Source: llm_adapter@2026-05-29

Confidence: high

Feature Low

Added current‑gen model pricing entries: Claude 4.x/4.5/4.6/4.7, Gemini 2.x/2.5, Grok, o1/o3, gpt-5.2; corrected `gemini-1.5-pro` rates and added `xai` provider

Added current‑gen model pricing entries: Claude 4.x/4.5/4.6/4.7, Gemini 2.x/2.5, Grok, o1/o3, gpt-5.2; corrected `gemini-1.5-pro` rates and added `xai` provider

Source: llm_adapter@2026-05-29

Confidence: high

Bugfix Critical

`formatMicrocents`, `set_budget`, and `get_task_roi` now correctly handle microcent conversion (100× off bug fixed)

`formatMicrocents`, `set_budget`, and `get_task_roi` now correctly handle microcent conversion (100× off bug fixed)

Source: llm_adapter@2026-05-29

Confidence: high

Full changelog

Highlights

🔴 Money bug — formatMicrocents / set_budget / get_task_roi were 100× off (correctness)

Convention is 1 cent = 10,000 microcents → 1 dollar = 1,000,000 microcents (source: agentledger apps/gateway/src/cost.ts). The 0.2.1 "100× too high" fix was compensating for a transient backend /100 inflation bug that was later fixed properly; the client compensation never reverted, so costs displayed 100× too low and budgets stored 100× too high (a $50 cap was being persisted as $5,000).

Live verified against real spend in prod — 3 high-spend agents reconcile to the cent under the new divisor (e.g. compliance dashboard $18.19total_microcents: 18,185,565 / 1,000,000 = $18.19 ✅; pre-fix this rendered as $0.18).

🟡 get_attribution_report rewired against the actual /api/outcomes shape

Tool was sending model and agent_id query params plus expecting a nested aggregate response that the endpoint never returned. Repointed to consume the flat outcomes list and aggregate (count, total value, top contributors) client-side. Enum updated to canonical ['direct','session_direct','first_touch','last_touch'].

🟡 get_task_roi date-range cross-version compat

Now dual-sends days + start_date/end_date so the tool works against the legacy backend (days-only) and the date-range-aware backend that shipped alongside.

🟡 Model pricing reconciled (May 2026)

Added current-gen models: Claude 4.x/4.5/4.6/4.7, Gemini 2.x/2.5, Grok, o1/o3, gpt-5.2. Corrected gemini-1.5-pro to $1.25 / $3.75 per 1M (Google price cut). Added xai provider.

🟢 Hygiene

Version sync across 7 locations. Experiment auto_promote copy fix. Regression pin in src/services/formatters.test.ts. CI workflow secret reference updated.


See #19 for full details and the live-smoke comment. Companion backend PR: ckpark123/Metrx#223. CI secret fix: #20.

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track metrxbots/mcp-server

Get notified when new releases ship.

Sign up free

About metrxbots/mcp-server

AI agent cost intelligence — track spend across providers, optimize model selection, manage budgets with enforcement, detect cost leaks, and prove ROI. 23 tools across 10 domains.

All releases →

Beta — feedback welcome: [email protected]