Prelint

Evidence checked September 12, 2026. Prelint’s live pricing metadata states $1 per completed review with no seats or subscription and $10 in free credits without a card. Its FAQ says purchased credits range from $10 to $1,000, do not expire, and failed, cancelled, or timed-out reviews are free. Auto-recharge defaults on when balance drops below $10, adding $20, but settings can change or disable it; monthly caps are available. The homepage explicitly says MCP is in closed tests. This file therefore no longer claims an available MCP server.

Prelint checks pull requests against specifications, compliance rules, business constraints, and a sourced decision ledger. Conventional linters inspect syntax; general AI reviewers focus on defects. Prelint targets product drift: code that may work technically but conflicts with what a team decided to build.

What it offers. Verified capabilities include specification-aware review, product decision ledger, GitHub integration, coding-agent CLI, coding-agent context (MCP remains in closed tests). Practical uses include preventing requirement drift, giving agents current decisions, reviewing small pull requests. The official price is $1 USD per completed review. Failed, cancelled, and timed-out reviews are free; credits from $10 to $1,000 do not expire. Auto recharge is on by default but can be changed, and monthly caps pause reviews. The FAQ says MCP is in closed tests, so general MCP availability is not verified. Relevant comparisons and guidance. Review CodeRabbit, GitHub Copilot Agents, and Cursor Agent, AI Coding Agents Comparison. These are comparison points or complementary products, not interchangeable substitutes. Buyers should map required inputs, actions, approvals, and exports before selecting a platform. Pricing and procurement. Current staged pricing is: Usage based: $1/completed review (failed or cancelled reviews free); Trial: $10 free credits (no card required) Public prices can change with billing cadence, usage, seats, support, and annual commitments. Where numeric pricing is absent or a pricing route is missing, treat the product as contact-sales and manually verify the quote. Ask for implementation, model or compute usage, overages, premium integrations, environments, support, data retention, and termination costs separately. Never infer that an unpublished price means a free tier. Practical pilot. Start with one bounded workflow and representative historical data. Define a baseline before connecting production systems. Test successful cases, ambiguous inputs, stale data, permission failures, upstream outages, and cases that should be rejected or escalated. Measure task completion, accuracy, false positives, false negatives, latency, operator review time, correction rate, and total operating cost. Require evidence or citations where the output depends on source records. A polished demonstration is not a substitute for a replay test using the buyer's own difficult cases. Security and operations. Apply least privilege to every connector and begin read-only where possible. Verify single sign-on, role-based access, audit history, encryption, retention, deletion, regional hosting, subprocessors, model providers, exportability, rate limits, incident procedures, and rollback. Consequential actions need explicit allowlists and accountable human approval until measured results support more autonomy. Test what happens when the service or an integration is unavailable; a production workflow needs a safe fallback rather than silent failure. Pros. $1 per completed review is easy to model; Failed and cancelled reviews are free; Credits do not expire; Targets product drift beyond code style. Cons. MCP is in closed tests, not generally available; Findings depend on current decisions; Charges add up with many small pull requests; Product judgments need accountable reviewers. Honest assessment. This tool is worth shortlisting when its specialized workflow is already a measured bottleneck and the organization can own integration, governance, and ongoing evaluation. It is a weaker fit when source data is poor, task volume is low, or nobody is accountable for exceptions. Calculate value from avoided labor, faster cycle time, reduced incidents, or improved outcomes, then subtract subscription, usage, implementation, training, and oversight. Expand only after the pilot shows durable gains without unacceptable errors. Decision checklist. Ask the vendor to demonstrate the exact connectors, data boundaries, failure handling, and administrative controls needed in production. Confirm which claims apply to the purchased plan and deployment region. Document who reviews outputs, who can change policy, how corrections are retained, and how the organization exits with its data. This discipline matters more than broad AI branding.