> ## Documentation Index > Fetch the complete documentation index at: https://docs.brilo.ai/llms.txt > Use this file to discover all available pages before exploring further. # Brilo MCP server > Let Claude Code, Cursor or your own AI agent read your Brilo agents and calls through the Brilo MCP server, signed in with an API key. The Brilo MCP server lets an AI tool such as Claude Code, Cursor or VS Code read your Brilo agents and call history, signed in with a Brilo API key. The tool can then answer questions like "how many calls went to voicemail yesterday" on its own. It is read only, and it only ever sees the workspace your API key belongs to. ## What it is The Brilo MCP server is a connection point at `https://api.brilo.ai/mcp` that speaks the Model Context Protocol (MCP), the open standard AI tools use to call outside services. An MCP client connects to it, asks which tools it offers, and calls them when a question needs your Brilo data. The tools list and read your agents and calls. The full list is on [MCP tools](/api-reference/mcp/tools). The Brilo MCP server is not the same as the documentation search that docs.brilo.ai offers to AI tools. That one searches these help pages. The Brilo MCP server reads your own workspace data. ## When you would use it The Brilo MCP server is useful whenever you want an AI tool to work with your real Brilo data instead of guessing: * **Reviewing calls in your editor.** A developer asks Claude Code which calls failed today and why, while fixing the action that caused it. * **Checking an agent's settings.** Someone asks Cursor whether recording and voicemail are switched on for the agent that answers the main line. * **Your own AI agent.** An internal agent you built reads last week's inbound calls to write a summary for your team. ## How it works The Brilo MCP server signs in every request with a Brilo API key, sent as `Authorization: Bearer YOUR_API_KEY`. The key decides which workspace the tools read, and no tool accepts a workspace as a parameter, so a client can never read another workspace. Every tool is read only. Each request is handled on its own, with no session to keep open. Your MCP client handles this; you only give it the server address and your API key. Each API key has its own rate limit, the one described in the [API introduction](/api-reference/introduction), so two clients on separate keys never slow each other down, even from the same office network. ## What is withheld The Brilo MCP server returns only the fields a customer needs, because whatever it returns goes into an AI model you chose rather than one Brilo runs: * **Transcripts and AI summaries** are returned for every call, except in HIPAA workspaces. * **HIPAA workspaces** never receive transcripts or AI summaries over MCP. * **Recordings** are never returned. They are kept in private storage that only your Brilo dashboard can open, so a link would not work in an outside tool. * **Internal details**, such as storage paths, internal ids and system error text, are never returned. Each tool lists the exact fields it returns on [MCP tools](/api-reference/mcp/tools). A HIPAA call with withheld content has `content_redacted` set to true and a `redaction_reason` of `hipaa`, so your client can tell it apart from a call with no transcript. ## What it costs and what it needs The Brilo MCP server needs a Brilo API key, from **Settings**, then **API Keys**, described in [Workspace settings](/account/workspace-settings). It also needs an MCP client that can send a request header, such as Claude Code, Cursor or VS Code. ## What else you can do here * [Connect an MCP client](/api-reference/mcp/connect) with copy and paste setup for each client * [See every MCP tool](/api-reference/mcp/tools) with its parameters and an example request * [Use the REST API](/api-reference/introduction) to start calls or change agents from code * [Send finished calls to your systems](/api-reference/webhooks/post-call) with a webhook ## FAQ An AI tool connected to the Brilo MCP server can read your agents and your call history: list agents, read one agent's full settings, list calls with filters such as date and direction, and read one call in full. It cannot change anything. The full list with parameters is on [MCP tools](/api-reference/mcp/tools). No. The Brilo MCP server reads only the workspace your API key belongs to. No tool takes a workspace as a parameter, so even a client that tries cannot reach another workspace. If you manage several workspaces, create a separate API key in each and connect them as separate servers in your client. Your workspace is most likely a HIPAA workspace. The Brilo MCP server never sends transcripts or AI summaries from HIPAA workspaces to an outside AI tool, because that content is protected health information. Every other workspace receives transcripts through the `get_call` tool. A call that was never answered has no transcript at all. The Brilo MCP server does not return recordings. Recordings are kept in private storage that only your Brilo dashboard can open, so a recording link would not play in an outside AI tool. To listen to a call, open it in your Brilo dashboard. The transcript and summary are available over MCP instead. No. The documentation site offers its own MCP connection that searches these help pages. The Brilo MCP server at `https://api.brilo.ai/mcp` reads your workspace data: agents and calls. You can connect both to the same AI tool. The docs one needs no key, the Brilo MCP server needs your API key. Not yet. The chat apps in the browser expect an OAuth sign in for outside servers, and the Brilo MCP server signs in with an API key today. Tools that let you set a request header work now, including Claude Code, Cursor, VS Code and AI agents you build yourself. Usually not. Connecting takes one command in Claude Code, or pasting a short block of settings into Cursor or VS Code, and takes about five minutes. You need an API key from **Settings**, then **API Keys**. The steps are on [Connect an MCP client](/api-reference/mcp/connect). ## Related * [Connect an MCP client](/api-reference/mcp/connect) to set it up in five minutes * [MCP tools](/api-reference/mcp/tools) for every tool and parameter * [API introduction](/api-reference/introduction) for API keys and rate limits * [Workspace settings](/account/workspace-settings) for where API keys live