Commercial Agent Tools: October 29, 2026

1. BugTape Records Production Errors as Evidence Coding Agents Read Over MCP

BugTape. BugTape, posted to Show HN on October 28, is error tracking built for coding agents: it groups a failure’s steps, failed request, console output and stack into one issue, and an agent such as Claude Code, Cursor, Codex, ChatGPT or Grok reads that evidence through an HTTP MCP endpoint and opens a pull request for review. Setup runs npx -y bugtape init in the app’s repository, with SDKs for the browser (reconstructed page-activity replay), Node.js, Python and a handled-errors-only Swift alpha. Inputs are masked by default, a strict mode also masks page text, and request and response bodies are not captured. Pricing: a free tier with 1 project, 100 new issues and 150 credits a month, Pro at $29 a month with unlimited projects and agent reads over MCP, and prepaid credits at $10 for 2,000. Source

2. Veqen Tracks Spend Across Coding Agents and AI Providers and Shuts Off Misused Keys

Veqen. Veqen, posted to Show HN on October 28, consolidates what a company pays across AI providers and work and coding agents (the site lists ChatGPT, Claude, Cursor, Copilot and Devin), attributing each charge to a person, customer or project. It flags, usually within the hour, when a key’s spend does not match anything the team or product made and, if enabled, shuts off only that key; it also reprices each day’s charges against the provider’s own rate card to find overcharges and produces month-end accounting entries. It connects to billing with read-only admin keys and does not see prompts, with SDKs via pip install veqen and npm i @veqen/sdk. Pricing: a free audit with no card required; paid plan pricing was not confirmed. Source

3. Arketum 1.1 Is an MCP Server That Gates Agent-Built Software Behind Human Decisions

Arketum. Arketum, from rlabs-cl, is an MCP server with a single build tool that a coding agent calls at every step; each response says where the work stands, what comes next and with what context. It runs no code and calls no model, but checks that what gets built matches what was asked, requires the person to confirm the plan before building, ties every change to a versioned requirement, and records human decisions with their provenance. It installs as a Claude Code plugin (with a local panel, a skill and three commands) or in any MCP client via npx -y @rlabs-cl/arketum, and is listed on the official MCP registry as cl.rlabs/arketum. Version 1.1.0 was published October 28 and posted to Show HN the same day. Pricing: free to use under a proprietary license that prohibits copying or modifying the code; telemetry can be disabled with ARKETUM_TELEMETRIA=0. Source