Model Context Protocol

Give Your AI Agent
Real Financial Data

PennyBot exposes a full MCP server — connect Claude, Cursor, or any MCP-compatible agent to your live transactions, accounts, and reports in minutes.

Start Free — Get API Key Read the Docs

Free for all users · No credit card required · 32 tools

Up and Running in 3 Minutes

Generate an API key in Settings → Developer, then add one block to your Claude config.

~/.claude.json

// Add to your Claude Code config
{
  "mcpServers": {
    "pennybot": {
      "type": "http",
      "url": "https://pennybot.io/mcp",
      "headers": {
        "Authorization": "Bearer pb_mcp_<your-key-here>"
      }
    }
  }
}

Create a free account

Generate an API key

Settings → Developer → Generate New API Key. Copy your key.

Add to your agent config

Paste the block above into ~/.claude.json with your key.

Start querying

Ask Claude "What did I spend on software last month?" — it just works.

31 Tools, Zero Setup

A complete MCP tool surface covering every financial data type in PennyBot. All tools return structured JSON — no scraping or parsing required.

Read Tools (14) — no write access needed

get_accounts

All connected accounts + balances

list_transactions

Filter by date, account, category

get_transaction_detail

Full metadata for one transaction

generate_expense_report

Spending by category + period

generate_net_worth_report

Assets, liabilities, net worth trend

get_user_profile

Account info, plan, preferences

list_connected_banks

Teller-connected institutions

create_session_note

Persist context across agent sessions

Write Tools (7) — requires read_write key

categorize_transaction

Set category + confidence

bulk_categorize

Up to 100 transactions at once

create_transaction

Add manual transactions

update_transaction

Edit description, notes, flags

bulk_flag

Flag/unflag multiple transactions

Batch + Import Tools (10) — stage changes, review, then commit

lock_records

Advisory lock for safe batch edits

commit_batch

Apply all pending changes atomically

preview_import

Dry-run CSV/JSON import with dedup

import_transactions

Bulk import up to 5,000 rows

find_duplicates

Smart fuzzy deduplication

See full tool reference →

Built for Agents, Not Just APIs

MCP is the right protocol for AI agents — structured tool calls, typed schemas, streaming responses.

🔒

Scoped Access

Create read-only keys for reporting agents and read_write keys for automation agents — independently revocable, never exposed again after creation.

📦

Batch Mode

Switch any key to batch mode: changes are staged in a pending queue and only committed when your agent explicitly calls commit_batch. Safe for bulk rewrites.

🗒️

Session Notes

Agents can persist context across sessions with create_session_note — store goals, decisions, or in-progress state that survives conversation resets.

⚡

Streamable HTTP

Standard MCP Streamable HTTP transport — works with Claude Code, any MCP client library, or a raw HTTP client. No SDK lock-in.

🛡️

User-Scoped Data

Every API key is tied to exactly one PennyBot user. Agents can only see data belonging to the key owner — row-level security enforced in the database.

📊

Real Financial Data

Transactions synced live from your bank via Teller, your accounting software, and manual entries — not synthetic examples or demo data.

What Agents Build With This

Real workflows PennyBot agents can handle end-to-end.

🤖

Monthly Close Agent

Pull all uncategorized transactions, categorize in bulk using AI reasoning, generate an expense report, and email it — all in one Claude conversation.

📈

Tax Prep Agent

Query all deductible transactions by category (travel, meals, software), flag unusual items for review, and export a structured summary for your accountant.

💬

Financial Q&A

"How much did I spend on contractors last quarter?" — Claude calls list_transactions with filters and answers from real data, not hallucinations.

🔄

Data Migration Agent

Import historical transactions from another tool, run preview_import to catch duplicates, review the diff, then import_transactions to commit.

Give Your AI AgentReal Financial Data