# How to connect Vaaya to Muse

_By Nakul Kelkar, September 28, 2026_

Connect Vaaya to Muse once, verify the connection, then give Muse one task with a budget. The first check should establish that the tools are available before you ask it to buy research, generate a file or start another paid job.

Vaaya's [Muse guide](/muse) and [agent-readable instructions](/muse.md) describe the current setup. This walkthrough follows that OAuth connection flow. It does not depend on copying a secret into the conversation or on a particular Muse menu label.

## 1. Start with a connection-only prompt

You need access to Muse and a Vaaya account. Reuse an existing Vaaya connection if Muse already has one. Otherwise, connect the remote MCP server at `https://vaaya.ai/mcp` and complete sign-in and OAuth approval yourself in the browser.

Paste this into Muse:

```text
Read https://vaaya.ai/muse.md and help me connect Vaaya.
Reuse my existing Vaaya connection if there is one.
Otherwise connect the remote MCP server at https://vaaya.ai/mcp.
Open https://vaaya.ai/signup if I need an account and let me
complete sign-in and OAuth approval in my browser.
Never ask me to paste a password, API key or token into chat.
Verify the connection with vaaya_test_connection.
For now, only connect and verify. Do not make a paid call,
purchase or trade. If a step fails, explain what failed.
```

The prompt stops after connection verification. Setup alone does not authorize a demonstration purchase.

## 2. Check the connection before planning work

Have Muse run `vaaya_test_connection` and report whether it succeeded. This check does not buy a service. If sign-in did not finish or a tool is unavailable, fix that connection step before planning paid work.

Do not use a screenshot of a settings screen as the only evidence. A connection entry can exist while authorization is incomplete. The tool response should establish whether Muse can reach the account now.

If your Muse environment cannot add or use the connection, preserve the failure message and consult the current [setup instructions](/muse). Avoid trying guessed endpoints or sharing credentials in an attempt to get past the error.

## 3. Learn the three calls used during a task

`consult` takes your goal and suggests suitable services and calls. It does not execute the proposed paid work. `use` runs a selected service action with its inputs and a `max_cost_cents` ceiling. `result` checks an asynchronous job returned by an earlier call.

Have Muse discover current schemas and prices through the [catalog](/api/catalog) or [tool reference](/docs/tools). A service being listed does not tell it how much of your budget it may spend. Give it a total allowance and have it track the remaining amount across calls.

For a multi-step job, ask for the sequence and cost estimates before execution. Save completed results so a later step does not purchase the same input again.

## 4. Run one small research task

A useful first task is to read one public page and extract a few facts. Choose something you can check yourself, such as your own pricing or documentation page. Use this separate prompt after setup:

```text
Use Vaaya to read [public URL] and return [three specific fields].
My total budget for this task is [amount]. Consult for the
current route, schema and price, then run within that budget.
Set a ceiling on each use call and track remaining spend.
Read this page only. Cite the source URL and retrieval date.
Mark absent fields unknown. If a call returns a job_id, poll
it with result rather than starting the work again.
Return the findings and actual charges. Stop when done.
```

Keep the first task read-only. Sending an email, making a call or changing another account is a separate action to specify when you need it.

## 5. Inspect the handoff and expand from there

Compare the returned facts with the page. Confirm that each citation supports the associated claim and that missing information stayed missing. A transport success alone does not establish that Muse read the intended page.

For an illustrative handoff, ask for: `source_url`, `retrieved_at`, `fields`, `unknowns`, `calls_used` and `actual_charges`. Those are fields for your report, not a claim that every provider returns that exact schema. If a charge or date is unavailable, have Muse say so.

Once that task works, choose a [Vaaya recipe](/recipes) for research, media or another workflow. Set the outcome and total budget each time, reuse previous results where appropriate, and keep the completed artifact with the call record that produced it.

## Questions

**Do I need to paste a Vaaya API key into Muse chat?**

No. The current Vaaya Muse guide uses the remote MCP server at https://vaaya.ai/mcp and browser-based sign-in and OAuth approval. Keep passwords, keys and tokens out of chat.

**Does checking the connection buy a service?**

No. vaaya_test_connection checks the account connection. The setup prompt in this tutorial stops after verification and does not authorize paid work.

**What should Muse do with an asynchronous job?**

Keep the returned job_id and check it with result. Do not resubmit use as a way to poll an existing paid job.