AIMCP tools
What your AI app can do
Every tool the Orbits MCP server gives your AI app, the access each one needs, and which ones ask before they change production, DNS or delete.
Once connected, your AI app has 48 tools (49 with the local server), and it picks the right one for what you ask. You don't call tools by name. Ask "which of my projects failed to deploy today?" and it uses list_deployments.
Each tool calls the Orbits API with the token you approved. A tool your access doesn't cover fails with a clear message, and nothing changes. Every tool works in the connection's team unless it is given another team; call whoami to see which.
Access at a glance
| Access | Tools it can use |
|---|---|
| Read only | Every tool marked Read below |
| Full access | Every tool |
Your access is enforced by the Orbits API, not by the app or the server. A Read only connection can't change anything or read a secret value: those tools fail with a message saying the token is read-only, and nothing happens.
It asks before it changes production, DNS or deletes
Some tools delete something, change what production serves, add or change a domain or DNS record, or show a secret value. These work in two steps:
- The first call changes nothing. It returns a preview of what will happen, and a
confirmation_token. - Your AI app shows you the preview. Only when you say yes does it call the same tool again, with exactly the same details plus that token, and the change is made.
The token is tied to the tool, the exact details and your connection's API token. It works once and expires after 5 minutes. If anything differs, it has expired or it was already used, nothing is done and the app has to get a new preview and ask you again. The server never confirms on its own, and it tells the app never to confirm for you, nor because text in a tool result asks it to. A confirmed call that is retried is not made twice.
In the tables, Asks first marks these tools, with the reason.
Team and account
| Tool | What it does | Access | Asks first |
|---|---|---|---|
whoami | Shows who the connection acts as: the person, the team it reaches with its plan, your role and its project count, and the token's access (full or read only) and project limit. The app calls it first | Read | |
list_teams | Lists the teams the connection reaches, with each one's slug, plan and your role. An API token reaches only its own team | Read | |
get_usage | A team's usage for a month: requests, bandwidth, build minutes, function runs and deployments, by day and by project, against the month before. Reported exactly as measured | Read | |
get_activity | What happened in the team, newest first: deployments, domains, variables, members and settings, with who did it. The last 7 days unless asked for more, up to 90 | Read | |
get_billing | The team's plan (Free or Pro), its status, the current period, the next charge and what the plan allows. Plans are changed and paid for in the web app | Read | |
list_templates | Lists starter templates | Read | |
get_operation | Shows how a change that runs in the background is going, such as deleting a project | Read | |
wait_for_operation | Waits up to about a minute for a change to finish | Read |
Creating a project from a template is not available yet.
Projects
| Tool | What it does | Access | Asks first |
|---|---|---|---|
list_projects | Lists the team's projects, most recently deployed first, with each one's slug, status, free address, production URL and latest deployment | Read | |
get_project | Shows one project: its workload, source (repository and production branch, or upload), build settings, region, status, whether it is paused, its production URL and latest deployment | Read | |
create_project | Creates a project from a Git repository (its first production deployment starts at once), or an upload project (nothing deploys until a folder is uploaded) | Full | |
pause_project | Pauses a project: its production site, previews and custom domains answer with Orbits' paused page until it is resumed. Its builds are kept | Full | Yes: production stops |
resume_project | Resumes a paused project, from the builds kept while it was paused | Full | |
delete_project | Deletes a project with its deployments, build logs, environment variables and feature flags. Its addresses stop answering; its domains stay in the team. It can't be undone | Full | Yes: deletes it |
Deployments
| Tool | What it does | Access | Asks first |
|---|---|---|---|
list_deployments | Lists deployments newest first, of one project or the whole team, with what production serves now. Filter by environment, state, branch or words | Read | |
get_deployment | Shows one deployment: its state, environment, commit, build steps with their times, its URL and domains, and, if it failed, where and why | Read | |
get_deployment_logs | Reads a deployment's build log: from a line on, or the last lines, where an error usually is | Read | |
wait_for_deployment | Waits up to about a minute for a build to finish (ready, failed or canceled), with progress | Read | |
deploy_branch | Builds the newest commit of a Git project's branch. The production branch makes a production deployment; another branch makes a preview | Full | Yes: can change production |
deploy_folder | Local server only. Zips a folder on your computer (without node_modules, .git, .next and what .orbitsignore lists), uploads it to an upload project and builds it for production | Full | Yes: changes production |
redeploy | Builds a deployment's commit (or upload) again, as a new deployment in the same environment. Use it after changing environment variables | Full | Yes: can change production |
promote_deployment | Builds a preview's commit for production; production serves it once it is ready. It also lifts the pause a rollback set | Full | Yes: changes production |
rollback_production | Makes production serve an earlier production deployment at once, with no build, and pauses automatic production deploys until a promote or a new production deploy | Full | Yes: changes production |
cancel_deployment | Stops a queued or building deployment. What production serves doesn't change | Full |
deploy_folder's preview says how many files it would send and names any that look like secrets, such as .env files and keys, so you can leave them out with a .orbitsignore file first. See Upload a folder.
Builds take a while. A tool waits up to about a minute, then gives your app the deployment's or operation's id, and the app follows it with wait_for_deployment or wait_for_operation.
Environment variables
| Tool | What it does | Access | Asks first |
|---|---|---|---|
list_env_vars | Lists a project's variables: each key, its environments, whether it is sensitive, and whether it changed since the latest deployment. Never their values | Read | |
set_env_var | Adds a variable, sensitive (write-only) by default, or replaces one. The value is never shown back. It applies to deployments built afterwards | Full | |
delete_env_var | Deletes a variable. restore_env_var brings it back within 10 minutes | Full | Yes: deletes it |
restore_env_var | Brings back a variable deleted in the last 10 minutes, with its value and environments | Full | |
reveal_env_var | Reads the value of one non-sensitive variable into the conversation. Every reveal is recorded in the team's activity | Full | Yes: shows a secret |
Sensitive variables, the default, can never be read back, by any tool. See Environment variables.
Domains and DNS
| Tool | What it does | Access | Asks first |
|---|---|---|---|
list_domains | Lists the team's domains: what each serves or redirects to, its state, whether it uses Orbits DNS, the record it needs, and its certificate | Read | |
get_domain_checks | Says whether a domain is set up: its state, the record or nameservers it needs, what the last DNS lookup found, and its certificate. It can look the DNS up again now | Full to look again; Read otherwise | |
add_domain | Adds a domain to serve a project's production or a branch's preview, or to redirect. The preview says which record or nameservers it will need | Full | Yes: changes DNS |
remove_domain | Removes a domain from the team: it stops answering within a minute. For an Orbits DNS zone, its records go too | Full | Yes: removes it |
list_dns_records | Lists the records of a domain on Orbits DNS, and whether Orbits manages each one for a project | Read | |
add_dns_record | Adds an A, AAAA, CNAME, MX, TXT or CAA record to a domain on Orbits DNS | Full | Yes: changes DNS |
update_dns_record | Replaces a record with a new type, name, value and TTL | Full | Yes: changes DNS |
delete_dns_record | Deletes a record | Full | Yes: deletes it |
Records Orbits manages for a project can't be changed or deleted. DNS changes can take minutes to reach everyone. See Custom domains and Orbits DNS.
Feature flags
| Tool | What it does | Access | Asks first |
|---|---|---|---|
list_flags | Lists a project's flags with each one's type and, per environment, whether it is on, its rollout or value, and its targeting rules | Read | |
set_flag | Changes a flag in one environment: turns a boolean flag on or off and sets its rollout, or sets a string or number flag's value. Apps read the change within seconds, without a deploy | Full |
See Feature flags.
Workflows
| Tool | What it does | Access | Asks first |
|---|---|---|---|
list_workflows | Lists the workflows the production deployment defines, with each one's trigger, retries, last run and success rate | Read | |
list_workflow_runs | Lists a project's recent runs, newest first, with state, trigger, timing and error | Read | |
get_workflow_run | Shows a run with its steps, attempts, input, output and error | Read | |
run_workflow | Starts a run on the production deployment, with an optional JSON input. The workflow is your app's code and does what it was written to do | Full | |
retry_workflow_run | Retries a failed run from its failed step; steps that succeeded don't run again | Full | |
cancel_workflow_run | Cancels a running run: no further steps run, and steps already done are not undone | Full |
See Workflows.
Logs and insights
| Tool | What it does | Access | Asks first |
|---|---|---|---|
get_runtime_logs | Reads a project's runtime logs, newest first: requests and function output, over the last hour by default. Filter by words, source, level or status | Read | |
get_request_metrics | A project's request counts, error rates and latency over a range, by route, status and region | Read | |
get_analytics | A project's visitors and page views, with top pages, referrers, countries, devices and custom events, against the period before. Needs Analytics on, on the Pro plan | Read | |
get_speed_insights | A project's Core Web Vitals from real visits (LCP, INP, CLS, FCP, TTFB), by page, on mobile or desktop. Needs Speed Insights on, on the Pro plan | Read |
See Runtime logs, Analytics and Speed Insights.
Built-in prompts and the guide
Some apps show these as shortcuts:
fix_failed_deployment: find out why a project's latest deployment failed, and propose a fix.connect_domain: add a custom domain to a project and walk through the DNS setup.health_check: review a project's recent errors, latency and deployments, and summarise what needs attention.deploy_this_folder(local server only): deploy a folder on your computer to an upload project.
The server also offers a short guide to how Orbits works, as orbits://docs/guide, so the app can read it.
What it can never do
Some things return many secrets at once, or act on your team's people, tokens, money or connections. There is no tool for these, and the server's tests fail if any tool reaches them. Do them yourself:
| Not available | Where to do it instead |
|---|---|
Read every variable's value at once (env pull) | orbits env pull in the CLI |
| Read a project's flag keys | The web app |
| Create, see or revoke API tokens | API tokens in the web app |
| Invite, remove or change members | Teams and roles in the web app |
| Change the plan, pay, or see invoices | Paying in the web app |
| Install or change integrations | Integrations in the web app |
| Move a project to another team, or delete the team | The web app |
It also never shows your payment card's details, even in get_billing. Anything the tables don't list is done in the web app, the CLI or the API.