The Agently MCP is live
by•
Quick one: we just shipped the Agently MCP.
Your Agently brain now plugs into any agent you already use, Claude Code, Claude Desktop, Cursor, anything that speaks MCP. Ask from there and it answers with your whole company in context instead of a blank slate.
Setup is three clicks:
Settings
API keys
New key, paste it into your agent
On Claude Code there's a one-line command to copy straight in.
That's it. Your stack, your data, in every agent you run.
Would love to hear what you connect it to first.
27 views
Replies
"Anything that speaks MCP" is the easy half. The part I'd want to know is what the server exposes: a handful of coarse tools, or one per capability? Coarse tools mean the model has to guess the right arguments and gets it wrong silently. Fine-grained ones flood the context — I've seen a single server eat a third of the window before the user typed anything.
And on the "whole company" side: does the MCP layer inherit the permissions of whoever connected it, or of the person asking? Those come apart the moment two people share a Claude Desktop config.
Agently
@ark_y_k Fair questions. On tools: there's basically two, read the brain and write to the brain. The model isn't choosing among a hundred provider actions, so there's nothing to guess wrong and nothing eating your context. The brain handles resolution itself underneath and drips the context needed per prompt.
On permissions: the MCP connects to your Claude (or any other provider) as you, scoped to your accounts access, so it acts as the person it's connected for. Two people on one shared config is two people sharing one login, that's a shared-password problem, not something the server decides.
Setup sounds painless. Genuine question before I'd paste a key into Claude Code, does the MCP scope what a given agent session can actually see, or is it all-or-nothing once the key's in? Whole-company context is great until a session that only needed one project can suddenly read everything else too.
Agently
@raffay_sajjad Great question. The brain resolves itself with relational extraction, you don't need to worry about anything besides getting the work done. The session trailing summary will be stored but wont effect your whole-company context