> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cfo.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# cfo.ai for AI agents

> For AI agents connected to cfo.ai through its MCP server, and the developers who set them up: what the server offers, how an agent shares the Model with Ari and the user, how to start, and every agent page.

The cfo.ai MCP server, at `https://api.cfo.ai/mcp`, gives an AI agent Ari's public product tools. With them an agent can read a company's financial Model and, when the user asks, change it: Pages, Table blocks, Canvas blocks, formulas, Variables, Dimensions, Dimension mappings, and Scenarios, along with the workspace's connections, Skills, Triggers, memory, team and access, workspace settings, and the changelog's undo and redo. An agent can also hand a question to Ari: `edit_chats` starts or continues an Ari chat and `inspect_chats` reads Ari's answer. [MCP tools and call shapes](/agents/mcp-tools.md) lists every tool. These pages describe those tools for the agent and for the developer who connects it.

## One Model for the user, Ari, and the agent

The agent works in the user's live workspace, signed in as that user. There is no separate copy of the Model for agents. Whatever the agent builds appears in the workspace right away, where the user can inspect the assumptions, change the result, or continue the work with Ari. Every change to the Model, such as a formula, a Table block, or a Page edit, is recorded in the Model's changelog under the user's name and can be undone. Actions outside the Model are not in that changelog, and undo cannot reverse them: changes to connections, Skills, Triggers, memory, and access, such as invitations and share links, and emails sent. See [Work safely](/agent-instructions.md#work-safely).

## How to start

1. Connect the client and verify access with the [Agent instructions](/agent-instructions.md). A person setting up their own tool can follow [Connect an AI tool](/guides/connect-an-ai-client.md) instead.
2. Read the skills the server's instructions name, through `inspect_skills`: `build-model` before you create or change model logic, Variables, Dimensions, or saved tables; `build-visualizations` before you add a Canvas block; and `import-model` before you rebuild a Model from a workbook or export.
3. Inspect the existing workspace before you create anything, and reuse its Variables, Dimensions, and Pages.

## Pages for agents

> **[Agent instructions](/agent-instructions.md)**
>
> Connect a client, get the user's browser approval, choose the workspace, and work safely.

> **[MCP tools and call shapes](/agents/mcp-tools.md)**
>
> Every published tool, the one-action-block call shape, workspace selection, results, and pagination.

> **[Canvas blocks for agents](/agents/canvas-blocks.md)**
>
> Create, patch, and render-check Canvas blocks, and name blocks in page edits.

> **[Table view grammar](/agents/table-view-grammar.md)**
>
> The breakdown notation for a table's rows and columns, and the typed layout fields that carry it.

### Formulas

> **[Formula authoring for agents](/agents/formulas-overview.md)**
>
> Which tool answers which formula question, and the rules every formula write follows.

> **[Read and explain formulas](/agents/read-formulas.md)**
>
> Read a Variable's saved formulas and the table values they produce, then explain them.

> **[Create and update formulas](/agents/write-formulas.md)**
>
> Write formulas through saved tables or on a Variable, with every request and receipt shape.

> **[Formula scope and formula ranges](/agents/formula-scope.md)**
>
> Conditions that say where a formula applies, and Actuals and Forecast ranges.

> **[Analyze saved values without changing the Model](/agents/analyze-without-saving.md)**
>
> Answer from saved tables, and know when a new result needs a saved row.

> **[Validate and verify formula writes](/agents/validate-and-verify.md)**
>
> Check a write before and after it saves, read its receipt, and confirm the numbers.

## Reading these pages

These pages are left out of the site search, so they do not crowd the answers customers look for. Every page has a Markdown version: add `.md` to its address, such as `https://docs.cfo.ai/agent-instructions.md`. [llms.txt](https://docs.cfo.ai/llms.txt) lists every page on the site.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.