Silvia MCP: Bring Your AI CFO Into the Agent You Already Use

Your general AI agent can write code, research a market, organize a project, or work through a long list of tasks. But when a decision turns financial, the agent usually reaches a hard limit: it does not know your accounts, holdings, transactions, documents, obligations, or financial history.

Today, we are introducing the Silvia Model Context Protocol server, or Silvia MCP. It lets supported AI agents ask Silvia questions inside the workflows where you already work.

Connect Silvia to Codex, Claude Code, Grok or Grok Bot, or Muse. Sign in to your Silvia account once through a secure authorization flow. Your agent can then call Silvia when it needs financial context, while Silvia handles the account-aware analysis.

This is not a generic finance plug-in and it is not a raw export of your financial life. The outside agent asks Silvia a question. Silvia runs the same analysis path used by Ask Silvia, with the same connected context, model routing, tools, plan limits, and account controls. Silvia returns the answer to the agent so it can continue the larger task.

What MCP changes

Model Context Protocol is an open standard for connecting AI applications to external tools and services. Instead of building a separate one-off integration for every agent, Silvia can publish a consistent set of tools through one secure endpoint.

The practical benefit is simple. You can keep working in the agent that fits the task and bring in Silvia only when the work needs your financial context.

Consider a few examples:

• You are using Codex to model a major purchase. Codex can ask Silvia to review the decision against current liquidity, investments, liabilities, and other known obligations.
• You are using Claude Code to build an internal business analysis. Claude Code can ask Silvia a follow-up question about the owner's connected household picture without asking you to rebuild it in a spreadsheet.
• You are delegating research to Grok Bot. The Bot can send the financial part of the question to Silvia, then use the answer in the rest of its workflow.
• You are using Muse to organize a broader personal project. Muse can ask Silvia for the financial context that belongs inside the plan.

MCP does not turn every agent into a financial system. It gives the agent a secure way to call the financial system that already understands you.

What the Silvia MCP actually exposes

The Silvia MCP currently exposes two tools.

ask_silvia

This tool sends a question to Silvia. It can start a new conversation or continue an existing Silvia conversation by passing its session_id.

Silvia can use the financial context available to the signed-in user, including connected accounts, holdings, transactions, documents, and supported email context. She can also use the analytical tools available inside the Silvia product. The answer reflects the same routing, limits, and account permissions that apply when the user works in Ask Silvia directly.

Most answers return in the original tool call. When deeper work takes longer, Silvia returns a session ID rather than forcing the outside agent to abandon the task.

get_silvia_answer

This read-only tool collects an answer that Silvia is still preparing. The outside agent passes the session ID from the original request. If the analysis is complete, the tool returns the result. If Silvia is still working, the agent can check again later.

This two-tool design keeps the interface small. The outside agent does not need to understand every account provider, data model, research tool, or analysis workflow inside Silvia. It only needs to know how to ask Silvia and how to collect a longer answer.

Why this is different from pasting financial data into a chat

Copying balances into a general chatbot creates a frozen snapshot. It can omit an account, become outdated quickly, and lose the relationship between ownership, liquidity, taxes, debt, and goals.

The Silvia MCP works differently in three important ways.

First, the connection is tied to your Silvia identity. The MCP client sends you through Silvia's OAuth sign-in, and the resulting access token must include the dedicated silvia:mcp scope. You do not paste a Silvia password or a permanent API key into the agent.

Second, the agent calls Silvia rather than receiving an unrestricted account feed. Silvia evaluates the request through the same product system used by Ask Silvia. The MCP connection does not create a second financial profile or a parallel copy of your accounts.

Third, conversations can continue. A returned session ID lets the agent ask a follow-up in the same Silvia conversation, so the analysis can build on earlier context instead of starting from zero.

You should still review important outputs. Connected context can improve relevance, but it does not eliminate incomplete source data, account sync delays, uncertain assumptions, or the need for qualified tax, legal, accounting, or investment advice.

Copy the Silvia MCP connection

The same remote endpoint works across supported MCP clients:

https://www.cfosilvia.com/mcp

The endpoint uses Streamable HTTP and OAuth. When your agent asks you to authenticate, sign in with the Silvia account whose financial context you want to use.

Install Silvia MCP in Codex

The fastest method is the Codex CLI:

codex mcp add silvia --url https://www.cfosilvia.com/mcp
codex mcp login silvia
codex mcp list

The first command adds the remote server. The second starts Silvia's OAuth sign-in. The third confirms that the server is configured.

If you prefer to edit the shared Codex configuration directly, add this to ~/.codex/config.toml:

[mcp_servers.silvia]
url = "https://www.cfosilvia.com/mcp"

Then run:

codex mcp login silvia

In the Codex desktop app or IDE extension, open Settings, choose MCP servers, select Add server, choose Streamable HTTP, and paste the Silvia MCP URL. Save, restart when prompted, then select Authenticate to complete sign-in.

After setup, type /mcp in Codex to confirm that Silvia is connected. Try a first request such as:

Ask Silvia to summarize the three financial issues that deserve my attention this week. Use my connected financial context and separate confirmed facts from assumptions.

Install Silvia MCP in Claude Code

Run this command:

claude mcp add --transport http --scope user silvia https://www.cfosilvia.com/mcp

Start Claude Code:

claude

Inside Claude Code, run:

/mcp

Choose silvia and follow the browser sign-in flow. The --scope user option makes the connection available to your Claude Code user across projects. If you only want it in one project, use a project-scoped MCP configuration instead.

You can verify the saved connection from the terminal:

claude mcp get silvia
claude mcp list

Then try:

Use Silvia to review whether this plan creates any liquidity, concentration, or debt-service issues for me. Do not assume missing financial details.

Install Silvia MCP in Grok and Grok Bot

For a personal Grok connection:

1. Open grok.com/connectors.
2. Select New Connector, then Custom.
3. Paste the Silvia MCP endpoint:

https://www.cfosilvia.com/mcp

1. Complete the Silvia sign-in and authorization flow.
2. Start a conversation and ask Grok to use Silvia for a financial question.

For a Grok Team Bot, open the Bot in the Grok Bot desktop app. In the Bot's information pane, go to Setup, open Plugins, and add a Custom MCP server using Remote HTTPS. Paste the same Silvia endpoint. Because Silvia uses OAuth, each teammate can sign in with their own Silvia account when the Bot first needs the connection. The Bot then uses that person's financial context, not the owner's.

A useful first instruction is:

When a task depends on my personal financial position, ask Silvia instead of guessing. Keep Silvia's confirmed facts separate from your own assumptions and tell me when professional review is appropriate.

Install Silvia MCP in Muse

Muse supports custom connectors, and Meta's current guidance allows you to ask Muse to create one. Copy this message into a Muse conversation:

Create a Custom Connector named Silvia using this remote Streamable HTTP MCP server:
https://www.cfosilvia.com/mcp

Use OAuth for sign-in. After connecting, show me the tools the server exposes.

Muse will guide you through the connection and place credentials in its secure credential flow. If your version of Muse exposes custom connectors directly in Settings > Connectors, you can add the same URL there. Product availability and connector controls can vary by account, region, plan, or workspace policy.

After connecting, try:

Ask Silvia whether the budget for this project fits my current financial position. Use the answer as context for the plan, but do not change any financial account or make a transaction.

Prompts that use the connection well

The best requests tell the general agent what role Silvia should play in the larger workflow.

Add financial context to a decision

I am considering this purchase. Ask Silvia to evaluate the effect on cash reserves, debt, portfolio concentration, and known near-term obligations. Then summarize the tradeoffs and list every assumption that still needs confirmation.

Review a plan without asking the general agent to guess

Before finalizing this plan, ask Silvia which parts conflict with my connected financial picture. Do not infer balances or holdings that Silvia does not confirm.

Continue a financial conversation

Continue the same Silvia conversation and ask which result changes most if the expected cost rises by 20%.

Separate household and business implications

Ask Silvia to separate the business implications from the household implications. Include liquidity, taxes, debt obligations, ownership concentration, and any missing data that would change the conclusion.

What Silvia MCP does not do

The MCP connection is not permission to move money, place trades, pay bills, or make irreversible decisions. The current public MCP interface is centered on asking Silvia for analysis and collecting the result.

It also does not remove the controls that already apply to your Silvia account. Plan limits, unavailable data, connection problems, and product access rules still apply. If an account is missing or stale in Silvia, the outside agent cannot make it current by calling the MCP server.

Finally, MCP does not make a general agent and an AI CFO interchangeable. The general agent can coordinate the larger workflow. Silvia supplies the financial context and financial analysis that the workflow needs. Keeping those roles distinct is the point.

Your financial context, available where work happens

Financial questions rarely arrive as isolated finance exercises. They appear while you are researching a purchase, planning a business project, reviewing a contract, building a model, or delegating a longer task.

Silvia MCP lets the agent handling that work ask your AI CFO for the part it should not guess.

Copy the endpoint, connect the agent you use, and sign in to Silvia:

https://www.cfosilvia.com/mcp

If you do not have a Silvia account yet, create one here, connect the accounts and assets you want Silvia to understand, then return to your agent to complete the MCP sign-in.

Educational information only. Silvia's outputs are not individualized tax, legal, accounting, or investment advice. Review material decisions and verify source data with the appropriate qualified professional.

Sources

• Silvia MCP protected resource metadata
• OpenAI: Connect Codex to an MCP server
• Anthropic: Connect Claude Code to tools via MCP
• xAI: Grok connectors and custom MCP connectors
• xAI: Grok Team Bots
• Meta: How Muse works with connectors
• Model Context Protocol