BLEND Localization MCP

npx add-mcp 'https://app.getblend.com/mcp'

Your AI agent can now order human translation.

Connect Claude, Codex, or any MCP client, to BLEND once.
From then on your agent quotes the job, places the order, tracks the linguist’s progress, and hands you the finished files — all without leaving the conversation.

01

Connect

Pick the path for the client you use.

Claude Desktop or Claude on the web:

1. Open Settings → Connectors.
2. Choose Add custom connector.
3. Name it BLEND, paste https://app.getblend.com/mcp as the endpoint, and click Add. Leave the OAuth fields (Client ID and Client Secret) empty — BLEND’s connector doesn’t need them.
4. Click Connect on the new BLEND row. Your browser opens BLEND — sign in with the account you already use.
5. Approve the connection. Claude comes back with BLEND’s tools available.

Claude Code, from the terminal:

claude mcp add --transport http blend https://app.getblend.com/mcp

Claude Code, from the terminal:

/mcp

/mcp opens a list of your connected servers. Select blend, choose Authenticate, and finish the login in the browser tab that opens. The connection is remembered — you won’t be asked again.

If you can already log into BLEND, you can connect. There’s no key to generate, no secret to paste, no config file to edit.

02

Verify

Ask your agent.

who am i on blend?

That’s the fastest proof the connection works. If it’s live, the agent replies with your BLEND account name and email — the same identity you use to log into the portal. If you get an error instead, the connection didn’t complete; go back and reconnect.

Ask for a translation the way you'd ask a colleague.

1. You ask, in the conversation you’re already having.
2. The agent prices it and shows you the quote before spending anything.
3. You approve. It orders.
4. It checks progress on its own schedule and tells you when the work lands.
5. The translated files arrive in the same conversation, ready to use.

Nothing about that sequence is scripted. The agent chains BLEND’s tools itself, in the order this specific job actually needs, and figures out that order from the tools’ own descriptions — not from a playbook you wrote for it.

Nothing about the translation gets faster. Everything around it disappears.

Human translation still takes hours to days. That part doesn’t change, and BLEND MCP was never built to pretend otherwise. What disappears is the busywork wrapped around it: copying content out of the tool you’re actually working in, opening a portal, picking a project type, uploading files, checking back, checking back again, downloading the result, and moving it back to where the work lives.

That hand-off was always the tax. You just paid it so often you stopped noticing.

Machine speed to order. Human quality to deliver.

Every order still goes to a real, vetted BLEND linguist with the right language pair and subject expertise — exactly like an order placed by hand in the portal. The agent doesn’t translate anything. That’s still a person’s job, on purpose.

Questions worth asking

Do I need a BLEND account already?

No. If you’re an existing customer, your balance, rates, and project history carry over. The agent is another way in, not a separate account. If you’re new to BLEND, create an account and top up credits first. Everything the agent orders shows up in your portal as a normal project either way.

What can my agent actually order?

Seven project types: Translation, Proofreading, Translation + Proofreading, Machine Translation, MTPE Full, Human AI, and Professional Translation + Editing. The headline says “order human translation” because that’s the shift worth naming, but the same connection covers AI and hybrid project types too.

What stops my agent from spending money without asking?

Tool descriptions instruct the model to get a quote before ordering, so cost is on screen before anything is charged. If you want a hard rule instead of a default, just tell your agent to confirm with you every time before placing an order and it will.

Which clients does this work with?

Any client that speaks MCP over HTTP: Claude Desktop, Claude on the web, and Claude Code today, with more as they add support.

What if my agent gets something wrong — like an invalid language pair?

BLEND’s errors are written for the model, not just a log file. Each one names what was wrong and which tool call fixes it, so the agent corrects itself without needing you to step in. Nothing is charged for a failed or guessed call — only ordering spends credits.