MCP server · Dev changes
Describe a change. Get back a reviewed pull request.
A cloud engineering agent from your assistant: it clones the repo, makes the smallest focused change, runs the project’s own tests, pushes, and opens a PR. Also triages issues, reviews diffs, and audits dependencies. Push-first; it refuses and explains the external fix when a request can’t be solved in code.
now
What it does
The Dev Changes server takes a described change and ships it as a reviewable pull request. It works in a clean clone, makes one focused change matching the repo’s conventions, secures the work by pushing first, then opens a PR with the test result. It also proposes changes without pushing, triages issues against the real code, reviews diffs, and audits dependencies for safe bumps.
It is built for engineering teams, indie developers and AI coding agents who want routine changes and maintenance handled as proper PRs. Real data only: it never fabricates a change, a PR URL, or a passing test, and when a request depends on missing credentials or a product decision it stops and explains the external fix rather than hacking around it.
Endpoint
Add this URL to your MCP client.
https://mcp.openhelm.ai/dev/mcpnpx mcp-remote https://mcp.openhelm.ai/dev/mcpTools
What your assistant can call.
Available on https://mcp.openhelm.ai/dev/mcp
open_pull_requestImplement a described change and open a PR (clone, change, test, push-first).
propose_changeAnalyse the repo and propose a diff/plan without pushing.
triage_issueReproduce/analyse an issue against the code; propose the smallest fix.
review_pull_requestReview a PR or the latest changes: bugs, security, missing tests, conventions.
dependency_auditAudit outdated/vulnerable deps; propose safe bumps (optionally open a PR).
check_resultRetrieve the finished result of a longer-running change.
Get started
Connect in under a minute.
- 1
Add the remote server
In ChatGPT, Claude, Cursor or any MCP client, add a remote connector pointing at https://mcp.openhelm.ai/dev/mcp. No install, no terminal, no Node setup. For clients that need a local bridge, run npx mcp-remote https://mcp.openhelm.ai/dev/mcp.
- 2
Connect your OpenHelm account
Click Connect and you are sent to app.openhelm.ai/connect to authorise via OAuth 2.1. Your encrypted OpenHelm API key is wrapped in a signed token, the server itself stores no secrets.
- 3
Add a card to start the trial
New accounts add a card to activate a 14-day free trial (you are not billed during the trial). This keeps the shared cloud compute protected from multi-account abuse.
- 4
Run tasks from your assistant
Ask your assistant in plain English. It calls the right tool, OpenHelm spins up an isolated cloud browser sandbox, and returns evidence-backed structured results, polling check_result for longer runs.
Want recurring dependency upkeep and housekeeping PRs? The Repo & Developer Changes strategy runs it on a schedule.
Set up scheduled repo maintenance→Common questions about the Dev Changes server
More servers
Explore the rest of the OpenHelm MCP suite.
Developers, engineering teams & power users
General-Purpose
A raw, unopinionated AI browser agent and run executor, straight from your assistant.
Investment analysts, VCs & corporate strategy teams
Deep Research
Buy-side research tools, SEC filings, earnings triage, comps, grounded in official sources.
Growth marketers, content directors & SEO specialists
SEO & GEO
Programmatic SEO plus GEO, rank on Google and audit how AI engines cite your brand.
Founders, sales & ops teams, and AI agents
Email Outreach
Send goal-directed emails and let the agent handle the whole reply thread.
Sales, support & operations teams · soon
Phone Reach-Out
Place and handle phone calls to accomplish a goal — coming soon.
Sales, support & operations teams · soon
WhatsApp Reach-Out
Reach contacts on WhatsApp and handle the reply thread — coming soon.
Operations, procurement & personal assistants · soon
Online Buying
Find and buy products online to a brief, with approval — coming soon.
Want it run for you instead?
The MCP servers put the tools in your assistant. If you'd rather we run the whole workflow as a managed service, book a 20-minute call and we'll scope it.