MCP Connector
Build with your AI agent. Run it on Replov.
Replov publishes a remote Model Context Protocol server. Connect it once and you can ask your agent to deploy the project you are building, check whether an app is healthy, read its logs to find a bug, ship a new version, or roll back a bad release — in plain language, without leaving the conversation. Connections, agent deployments and agent activity are on the AI Agents page.
Inspect
Status, logs, deploy history
Deploy
Ship and roll back
Guarded
Destructive actions confirm twice
Scoped
One account, revocable
Connecting
In Claude, open Settings → Connectors → Add custom connector and enter:
https://mcp.replov.com/mcp
/mcp.Claude detects the rest automatically — choose Sign in now for authentication and Register automatically for the OAuth client. You will be sent to Replov to sign in and approve the permissions, then returned to Claude. No API key to copy, and your password is never shared with the assistant.
Two ways to deploy
From a workspace — no Git needed
deploy_source → short-lived upload → deployment
Ask “Deploy this project to Replov.” The agent gets a single-use upload link that expires in 10 minutes, zips the project and uploads it straight to Replov — the code never passes through the conversation. Replov builds it with its normal pipeline and returns the HTTPS URL. Works for web services and static sites. To ship a new version, ask again: the same app is updated.
From a Git repository
create_app → deploy_app
Ask “Deploy github.com/me/my-app to Replov.” The agent creates the app from the repository and branch, then deploys the latest commit. Later deploys rebuild the branch.
Both create apps on the Nano compute tier by default. Nano is billed per service at ₹199 per 30 days, except for the first two Nano Static Sites on an account. Both end as a normal Replov app — it keeps running after the conversation, with logs, deployment history and rollback in the dashboard, where agent deployments carry an AI agent badge.
What it can do
Read-only
| Tool | Ask it something like |
|---|---|
list_apps | “List my Replov apps” |
get_app_status | “Is my API up? What tier is it on?” |
get_logs | “Show me the last 100 log lines for the worker” |
get_deployment_history | “When did I last deploy, and did it succeed?” |
Write
| Tool | What it does |
|---|---|
create_app | Creates an app from a git repo. Defaults to the per-service Nano tier (₹199 per 30 days). Shows the price and your wallet balance and waits for your approval before creating a billed app. It can also create a cron job: a command run on a schedule in a timezone you choose; your agent then sees each run's status and exit code, can read a run's output and can start a run now. |
deploy_app | Builds and ships the latest commit of a Git app. |
deploy_source | Deploys a local workspace without Git: returns a single-use upload link (10 minutes) for a zip of the project, creating the app on first use at the per-service Nano tier, after you approve the price. It can create a web service, static site, background worker or private service; workers and private services get no public URL, and other apps in the same project reach a private service at its name and port. A folder of a multi-app workspace can be chosen with root_directory. Your first two Nano static sites are free and need no approval. |
scale_app | Changes the compute tier. Every tier is billed per app; moving to a more expensive one asks for approval first. |
rollback_deploy | Restores an earlier deployment. |
delete_app | Permanently deletes an app. Irreversible. Deleting an app leaves its Redis running (and billed) until you delete the Redis too. |
configure_app | Declares the environment variables an app needs and sets non-secret values. Secret values are never accepted: the result links to your dashboard, where you enter them, and the agent then sees them as configured without the value. |
provision_resource | Adds a Redis (Micro or larger) to an app. The first call only returns the exact price; the Redis is created when the call is repeated with the quote's single-use confirmation token after you approve. |
manage_domain | Connects a domain you own to a web or static app on a paid tier. The agent gives you a CNAME and a TXT record to add at your DNS provider (it never needs your DNS login), verifies them, and reports when HTTPS is live. Removing a domain needs a second, confirmed call. On a team, only the owner or an admin can change domains. |
Adding Redis
Ask your agent to add Redis to an application. Replov will show the exact price and ask for approval before creating the paid resource. After you approve, Replov creates the Redis in the app's project, connects it over the private network and configures REDIS_URL for the app. The connection string and password are never shown to the agent. Redeploy the app to apply it.
Destructive actions confirm twice
Deleting an app and rolling back a deployment do nothing on the first request, and neither does creating a billed app or moving an app to a more expensive tier. The connector describes exactly what would happen and stops. Only a second, explicit confirmation carries it out — and deleting additionally requires the app's name to be repeated back, so a deletion cannot be confirmed without the assistant having identified the specific app.
So asking to delete an app gets you a summary first:
⚠️ Confirmation required — nothing has been changed yet.Requested action: Permanently delete the app "api-gateway"What will happen:- The web "api-gateway" is stopped and removed from the cluster.- Its deployment history, environment variables and build cache are deleted.- https://api-gateway-a1b2.replov.com stops resolving.- Billing for this app stops.This cannot be undone.
Access and privacy
The connector can never do more than your own account can. It cannot reach your payment methods, wallet balance, account settings or password, and cannot see other users' resources. Access tokens expire after an hour, refresh automatically, and are revoked the moment you disconnect — from Claude, or from Replov under AI Agents.
deploy_source lists the names of any variables the project expects but the app does not have yet — names only, never values.Every action the connector takes on your behalf is logged, so you can review what an assistant did on your account under Agent activity on the AI Agents page.
Full detail in the MCP Connector Privacy Policy.
Troubleshooting
| Symptom | Cause |
|---|---|
| Won't connect | The URL is missing the /mcp path. |
| Connects, but no write tools | Only read permissions were approved. Disconnect and reconnect to approve the rest. |
| “Too many write operations” | A per-account rate limit on writes. Wait a minute — it exists so an automated retry loop cannot repeatedly deploy or delete. |
| “Upload token expired” or “already used” | Upload links work once and last 10 minutes. Ask the agent to deploy again for a fresh one. |
| “Insufficient wallet balance” | The affected paid service enters a 3-day grace period. Add wallet balance before the deadline to avoid suspension. |
| A tool says an app doesn't exist | The assistant used a stale id. Ask it to list your apps again. |