aldena mcp server: run your teams and agents from claude, cursor or vs code
connect claude, cursor, vs code or any other mcp client to mcp.aldena.ai/mcp with oauth or an api key, and run your teams, projects, agents and chats from it.
an mcp server is a service that publishes tools over the model context protocol, so an ai client can call them for you.
aldena runs one at https://mcp.aldena.ai/mcp. add that url to the client you already work in, claude, claude code, cursor, vs code or any other client that speaks mcp over http, and manage aldena from there.
it is the other direction from mcp servers in a project, where your agents call tools on servers you add. here, your own ai client calls aldena.
what a connection can do
a connection gets the tools the concierge uses, except the ones for api keys, plus tools from the rest api for teams, provider keys, shared agents, skill versions, repo imports and answering your agents.
that covers teams, projects and their settings, agents, skills, chats and connectors. every tool carries a title, an input schema, and hints that tell the client whether it only reads or can delete something.
api keys are managed in aldena only. no connection, oauth or api key, full or read-only, can list, rename, disable or delete a key: do that under api keys in the account menu, or ask the concierge in the app.
a connection acts as you in every team you belong to, with the role you hold in each. an admin-only tool refuses it in a team where you are a member, the same way the concierge refuses you.
a read-only connection never sees a write tool. if a client calls one by name anyway, it gets an error back and nothing runs.
connecting with oauth
add the url to your client. the client registers itself with aldena and opens a consent page in your browser.
the page names the app, shows where it will return to and which account you are signed in with, and asks for Full access or Read-only. press Allow and the browser hands you back to the client.
if you are signed out, sign in first, with a password, a magic link or google, or create an account, and you land back on the same consent page. an account has to verify its email before it can approve a connection.
a read-only approval stays read-only. its tokens refresh without asking you again, and no refresh ever adds write access.
connecting with an api key
a client can send a personal api key instead, as Authorization: Bearer ald_…. that suits a client without a browser sign-in, or a connection you want to hand a specific key.
the key's own access decides what the connection can do, so a read-only key lists only read tools. keys come from api keys in the account menu, work over mcp for as long as they are enabled, and a disabled or deleted key is refused from its next request.
setup for each client
open MCP in the account menu, just below Account. the page shows the server url with a copy button and a tab for each client, every snippet with its own copy button:
- claude code: one
claude mcp addcommand, and a variant that adds an api key header. - claude: the url to add as a custom connector under Customize, then Connectors. claude connects through oauth only.
- cursor: an entry for
~/.cursor/mcp.json, with or without the key header. - vs code: a config entry, with or without the key header.
- other: the url and the header line, for any client that takes them.
connected apps
the same page lists every live oauth connection with the app's name, its access, when it connected and when it was last used.
disconnect one and the app loses access on its next request. connecting it again shows the consent page again.
an oauth connection also ends on its own after 30 days without use, and drops off the list. api keys do not appear here: rename, disable or delete them under api keys.
the concierge can list your connections and disconnect one after you approve. it never approves a new connection or raises one to full access, and sends you to the MCP page instead.
limits and permissions
mcp calls count against the same per-minute limits as the rest api, per api key or per oauth connection, and the best plan among your teams scales them the same way. a key's mcp and rest calls share one set of limits.
a call over a limit gets a 429 with Retry-After, and runs nothing.
the allow, ask and deny policies you set for concierge tools under your account's permissions do not apply over mcp. your client asks you to confirm tool calls instead.
faq
which ai clients work with it?
any client that speaks mcp over streamable http. the mcp page has setup for claude code, claude, cursor and vs code, and clients on the 2025 protocol and on the 2026-07-28 protocol can both list and call tools.
what is the difference between full access and read-only?
full access lists every tool. read-only lists only the tools that read, and a write tool called by name returns an error without running.
is the mcp server on every plan?
yes. the plan only sets the rate limits, the same numbers the rest api uses.
can a connection reach only one team?
no. a connection acts as you in every team you belong to, with your role in each, and loses a team as soon as you leave it.
how long does a connection last?
an oauth connection lasts until it goes 30 days without use, or until you disconnect it. an api key works for as long as the key is enabled.
does the server offer resources or prompts?
no. it offers tools only.
spin up your first project.
one project per client, product, or piece of work, staffed with intake, requirements, build and code review agents.