v2 2× features

Connect via MCP. We build & maintain your API test suite.

TestingForge connects to your AI agent via MCP, generates a structured API test suite, runs it in CI, and proposes updates whenever your API changes — with a human reviewer in the loop.

API-first. No browser/UI tests. No empty collections to fill in by hand.

Product · Release notes

Release notes.

How TestingForge grew from a run-and-report tool into an MCP-native platform where an agent builds and maintains your API suite for you. Everything below actually shipped — here is the trail.

Latest v2.7 12 releases since April 2025 Updated July 6, 2026
v2.7 Runs Latest

Runs that never hang — per-brick timeouts & leaner memory

Made a single slow or unresponsive endpoint a non-event: every brick now runs under a hard time budget, memory is reclaimed as the suite progresses, and a tenant can run two suites at once.

  • Hard per-brick timeout: a brick that does not respond within its budget (default 60s, configurable per organization) is failed and the run moves on, instead of parking the worker for minutes.
  • The timeout bounds the whole brick, not just one call — retry loops and cleanup loops stop as soon as the budget is spent, and polling bricks are capped so a mis-set poll can no longer run for hours.
  • Memory is reclaimed after every brick (cycle collection plus closing response streams immediately) so long suites with large payloads stay well clear of the out-of-memory edge that used to kill a worker mid-run.
  • A single HTTP client is reused for the whole run instead of building a fresh one per request, cutting allocation churn and leaked-handle debris.
  • Per-organization concurrency raised so a tenant can run two suites at the same time.
v2.6 MCP

The MCP agent release — leaner reads, safe edits, reliable writes

The big one: the whole MCP tool surface reworked so an AI agent can drive it end to end — lighter payloads, safe partial edits, guaranteed cleanup and rock-solid writes on freshly created workflows. Built through late June and rolled out together on July 4.

  • Read-only brick views (compact / raw) that return a lean payload and never accidentally fork a template brick.
  • Field projection with friendly aliases — ask for just id, name, method, url, enabled, checks or extracts.
  • Safe partial edits: append a single check, header, extract or condition, or upsert a header by name, without resending the whole array.
  • Teardown groups (always-run) that execute cleanup even when an earlier step fails, and survive export/import and template forks.
  • Create empty workflows, dry-run against a custom target host, and try a single group in isolation.
  • Organization-aware scenario graph and full-text brick search across label, URL, method, description, tags and group name.
  • Leaner MCP tool responses so agents spend fewer tokens per call, and brick copies preserved correctly when cloning workflows.
  • A workflow created via the API is instantly editable — add groups and bricks, reorder, bulk-move, try, set as default, run or delete with no delay and no spurious 403.
  • Bulk move reports skipped items explicitly instead of returning a silent zero. Expanded MCP guides, manifest and schemas.
v2.3 Runs

Run reliability at scale

Hardened the run queue so heavy, slow or misbehaving targets can no longer take down execution.

  • Response body reads are capped so a huge response can no longer exhaust memory and kill the queue worker — with the cap tuned so large but valid bodies still parse.
  • Concurrent executions per organization are capped and serialized to protect shared target environments.
  • Runs left stuck in "running" after a worker crash are automatically reaped.
  • Live-progress updates no longer risk OOM-ing the worker; oversized check values are truncated in traces.
  • Old run logs are pruned daily, and a slow-connection delay between requests avoids tripping 503 rate limits.
v2.2 Performance

MCP concurrency & a much faster workspace

Kept the hosted MCP server stable under load and made the bricks workspace feel instant.

  • A Redis-based concurrency limiter on the hosted MCP server prevents worker exhaustion, and degrades gracefully when Redis is unavailable.
  • The bricks workspace now renders in under 100ms instead of the previous 5–8 seconds.
v2.1 MCP

MCP hardening & MCP-first product

Made the Machine API trustworthy and repositioned the whole product around the MCP workflow.

  • Added an MCP manifest validator, audit logging for tool calls and end-to-end test coverage of the Machine API.
  • Fixed argument handling so request-body fields that share a name with a path parameter are no longer stripped.
  • Reframed the landing pages and documentation around the MCP-first, connect-and-let-the-agent-work experience.
v2.0 Major

Machine API & Cursor MCP integration

TestingForge opened up to AI agents: a full Machine API and a hosted MCP server you connect to Cursor in a couple of lines.

  • New Machine API under /api/v1 covering bricks, workflows and the workspace.
  • First-class Cursor MCP integration with tools for bricks, workspace and source-control operations.
  • A hosted PHP MCP server at /mcp/server for copy-paste setup — no local binary to install.
v1.3 Docs

Documentation & the Bricks user guide

Wrote the docs so new teams can get from zero to a running suite on their own.

  • A dedicated documentation site with a step-by-step Bricks and workflows guide.
  • Coverage of CI usage with an API key, environments, secrets, runs and troubleshooting.
v1.2 Stability

Stability & regression suites

Shipped a large real-world regression suite and eliminated the memory blowups it surfaced.

  • Added a full regression suite driven against a real CMS API.
  • Fixed several out-of-memory conditions across the run pipeline, the run timeline and the bricks page.
v1.1 Runs

Live run controls

Made watching and re-driving a run far more direct.

  • A current-brick indicator on the live run page so you can see exactly where a run is.
  • Configurable poll limits for polling bricks (http_poll_until).
  • Start a new run or re-run straight from a brick’s actions, and longer brick labels.
v1.0 Major

Bricks & Pulse

The product took its current shape: editable API building blocks, an instant load-feel tool, and a public demo anyone can try.

  • The Bricks engine — every API call is a first-class, editable record with variables, assertions and revisions, arranged into drag-and-drop scenarios.
  • Pulse — fire any endpoint N times with a slow-response threshold and watch each request land as a live tile.
  • A public demo workspace that resets every hour, with the full merged catalogue of template and organization bricks.
  • PAT-based source-control connection, and polling that fails fast on 401/403 instead of hammering for minutes.
v0.4 Platform

Runs, reporting & billing

Turned the foundation into a usable service: run things, see results, and pay for it.

  • Queue-based run execution on Octane with a re-run mechanism and detailed run views.
  • Dynamic reports, extended logs, a dashboard and Slack report delivery.
  • Stripe billing with metered usage and invoices, plus team invites and onboarding.
v0.1 Launch

Foundation

The first cut of TestingForge went live.

  • Marketing site, user accounts and organizations, and transactional emails.
  • CI/CD pipelines, health checks and the core application skeleton the rest of the platform grew on.

Want the newest bits in your suite?

Connect the MCP and let the agent build and maintain your API tests — or take the live demo for a spin first.

Try the live demo →