Start free
Agent-native finance · v1 Beta · 2026

Your finance system, with agents in the room.

FinButler is a finance workspace built for the era after dashboards. People, accountants, and AI agents share the same numbers, the same permissions, and the same workflows — through a hosted MCP server that plugs into Claude, ChatGPT, and the tools you already use.

Start free in the dashboard Read the agent + MCP brief
Workspaces
Team + firm + client
MCP transport
streamable‑http
Permissions
Scoped, audited
Workspace · Northline Advisory
fin.mcp · live

  1. 10:02 · Claude · firm reviewer

    finbutler.list_open_reviews(client="Paper Oak Studio")

    3 transactions over $400 need a category. May proceed.

  2. 10:03 · Agent · auto‑categorise

    Suggested categories. Held for human approval — client review room only.

  3. 10:04 · Erick · firm admin

    Approved 2 of 3. Flagged 1 for client.

  4. 10:05 · MCP · audit log

    Written to workspace.audit. Permissions: shared‑with‑firm.

Scope: client · approver Trust by design
Hosted MCP https://mcp.finbutler.ai/mcp
The platform, flat

Three layers, one workspace.

We didn’t bolt an AI tab onto a CRUD app. Workspaces, agents, and MCP are the platform — designed so a person and an assistant can pick up the same task without anyone losing context or trust.

  1. 01 Layer

    Workspaces — people, firms, clients

    The same workspace model your team uses powers every other layer. Internal staff, accountants, and clients each get scoped access, not a free-for-all login.

    • Personal, team, firm, and client workspaces
    • Role-based access: owner, admin, member, viewer
    • Audit-ready logs on every action
  2. 02 Layer

    Agents — built in, not bolted on

    Categorisation, OCR, recurring detection, and review handoff are agent tasks under the same permissions humans have. Nothing is auto-applied without an approval rule you wrote.

    • Auto-categorise with held-for-review thresholds
    • OCR + structured invoice extraction
    • Hindsight loop for accuracy over time
  3. 03 Layer

    MCP — your assistant joins the workspace

    Plug Claude, ChatGPT, or any MCP-aware client into a hosted endpoint. The assistant sees only what the user can see — no copy-pasting bank data into a chat.

    • Hosted streamable-http endpoint
    • Workspace-scoped tools and resources
    • Same permissions as the rest of FinButler
Agent fabric

Built to be useful with an assistant in the room.

Every product decision starts with the question, “Can a human and an agent both do this safely?” That changes how categories work, how permissions work, and how reports get generated and shared.

Workspace fabric

Person

UI, mobile

Agent

in‑product

Assistant

via MCP

workspace.policy single source of truth
One policy fabric, three surfaces. Nothing routes around it.

  • Same data, same rules

    An agent reading your transactions is bound by the same workspace and role you’d give a teammate. No shadow integrations.

  • Held-for-review by default

    Agents propose; people approve. You decide where the threshold drops to “auto.” Every change is reversible and logged.

  • No copy-paste of finance data

    Your assistant queries FinButler over MCP. Numbers never leave the workspace, and prompts don’t become a second source of truth.

  • Hindsight, not hallucination

    Categorisation models learn from your historical decisions. New users get sensible defaults; long-time users get a model that fits them.

Bring your assistant

Works with the tools you already trust.

FinButler isn’t betting on a single model or a single chat product. It speaks Model Context Protocol, so the same workspace shows up inside Claude, ChatGPT, or whichever agent you decide is best for the job.

Read the MCP brief →

  • Claude

    Anthropic · MCP-native

    Add the FinButler MCP server in Claude Desktop or Claude Code and ask in plain language.

  • ChatGPT

    OpenAI · custom GPT + connector

    Connect through the FinButler GPT or your own MCP-aware agent runtime. Same scoped permissions.

  • Cursor / Continue

    Dev assistants

    Helpful for ops engineering — finance answers without leaving the editor.

  • Bank scrapers

    Built-in · DGI Panama, more on the way

    Pull QR invoices and statements directly into the workspace — agent or human can trigger.

  • OCR

    Gemini-powered

    Photograph an invoice or receipt; FinButler structures it before it ever hits the ledger.

  • Your tools

    MCP open spec

    Anything that speaks Model Context Protocol can read or write — under your permissions.

Hosted MCP server

One endpoint. The whole workspace.

FinButler ships with a hosted Model Context Protocol server. Point any MCP-aware client at it, sign in once, and your assistant can read accounts, propose categories, draft reports, or invite a client — inside the same permissions you’d give a teammate.

Transport

streamable-http

Auth

OAuth · workspace-scoped

Endpoint

https://mcp.finbutler.ai/mcp

Get MCP access Read the engineering notes
~/.config/mcp/finbutler.json json 
{
  "mcpServers": {
    "finbutler": {
      "transport": "streamable-http",
      "url": "https://mcp.finbutler.ai/mcp",
      "auth": "oauth"
    }
  }
}
Tool Use
list_accounts Pull every account a user can see.
list_transactions Date range, account, category, paid-by member.
categorize_pending Run the auto-categorise agent with held-for-review.
create_budget_review Draft a budget review the team can approve.
invite_client Pull a client into the right firm workspace.
export_report Produce a shareable report for accountants or banks.
Built for firms

A dedicated dashboard for accountants — without leaving the workspace.

Accountants get a firm-shaped surface for clients, invitations, managed teams, tax reports, and firm settings. Same workspace model, same permissions, same agents and MCP — different lens.

  • Clients

    Open a client, invite them into the right workspace, and keep every request tied to a single context. Approval rules are visible to both sides.

  • Teams

    Managed teams keep firm ownership clean. Move a client between senior, bookkeeper, and reviewer without scattering files.

  • Reports

    Tax and review packages get drafted inside the workspace, with the same audit trail clients see. Agents can pre-fill; humans sign off.

See the accountants page Open the firm workspace →

Firm workspace

Northline Advisory

14 active clients

Reviews this week

23

Held for approval

7

  • Paper Oak Studio

    Review pending · 3 items

    Review

  • Monarch Foods

    Active · agents on category 80% / hold 20%

    Active

  • East Harbor LLC

    Waiting on bank scrape

    Waiting
Pricing

Free for individuals. Real work pricing for teams.

Live prices and limits come from the product. The mobile and web plans pages always reflect what your account is on — these are the shapes, not the receipts.

Full pricing + feature table →

  1. Free

    For individuals putting their numbers in one place.

    • 1 workspace
    • Manual import + OCR sampling
    • Read-only MCP for personal use
    Start free

  2. Pro

    Most chosen

    For households and operators who want agents doing real work.

    • Agents on by default (held-for-review)
    • Full MCP access for one user
    • Recurring + budget intelligence
    Try Pro

  3. Unlimited

    For teams and accountants who share the same workspace.

    • Shared MCP for everyone in the team
    • Firm + client workspaces, scoped roles
    • Priority OCR + scraper bandwidth
    Talk to us

Need firm-grade controls? Talk to us about a Firm plan — custom limits, SSO, and dedicated support for accounting firms with 15+ clients.

Notes

Writing from the build, not from the brochure.

Engineering notes, product decisions, and finance-side observations. New posts land here when we have something specific to say.

Read every post →
  1. product ·

    Why we put MCP at the center of FinButler

    A short, candid memo on the decision to make agents and the Model Context Protocol first-class — and what changed in the product the day we stopped treating them as a feature.

    Erick Agrazal · Founder · 7 min read

FAQ

The questions you’d ask if we were in the room.

Short answers. Email hello@finbutler.ai if there’s something we missed.

Is FinButler safe to connect to my bank or my accountant?
Workspaces have role-based access, audit logs on every change, and a strict permissions model. Agents read what you’d let a teammate read. Sharing data with an accountant is opt-in, scoped to a single workspace, and revocable.
What does “agent-centric” actually mean here?
We design every feature so a human and an agent can do it under the same rules. Categorisation, OCR, reviews, even invitations are tools that an in-product agent or an external assistant (via MCP) can call — always under your permissions and held-for-review by default.
Why MCP and not a custom API?
We still ship a REST API. MCP is the part that lets Claude, ChatGPT, and other assistants share the same workspace without us shipping a separate plugin to each. One server, every agent.
Do my prompts or numbers train an external model?
No. Your data stays in the workspace. Assistants query it through MCP at request time, scoped to what you’re allowed to see. We don’t hand finance data to third-party training pipelines.
Is this ready to use for a small business?
Yes for early teams who want to set up workspaces, run agents, and use MCP. We’re still expanding bank coverage and adding firm-grade controls — Pro is the right place to start; Unlimited and Firm unlock the multi-seat workflow.

Get started

Bring your team, your accountant, and your assistant into one finance workspace.

Start free on the dashboard. Turn on agents when you’re ready. Add your MCP-aware assistant when it has real work to do.

Start free See the firm workspace
Hosted MCP endpoint https://mcp.finbutler.ai/mcp