Give AI Agents portfolio context with the Parqet MCP

What is the Parqet MCP?

The Parqet MCP server is a remote Model Context Protocol server that lets AI agents and other MCP-compatible clients read your portfolio data and add new data. It acts as a bridge between your Parqet account and AI, so you can ask questions about your holdings, performance, transactions, and dividends or ask the agent to create portfolios, holdings, and activities.

The server is hosted at https://mcp.parqet.com and uses OAuth 2.0 via Parqet Connect for authentication. Read and write access use separate OAuth scopes, so write tools only work when you grant permission to update portfolio data.

Write support is limited to creating portfolios and holdings and adding activities. The MCP does not edit or delete existing portfolio data.

Key features

• Portfolio overview: list all your portfolios with names, currencies, and connected brokers
• Portfolio management: create portfolios and add cash, custom, real estate, insurance, P2P, or commodity holdings
• Holdings & valuation: see current positions, market values, weights, and total portfolio worth
• Performance metrics: get XIRR, TTWROR, realized/unrealized gains, dividends, fees, taxes, and net allocations (net exposure)
• Activities: browse or add buys, sells, dividends, transfers, deposits, withdrawals, fees, and taxes
• Multi-portfolio aggregation: query performance across multiple portfolios at once

Authentication

The Parqet MCP server uses OAuth 2.0 authorization code flow via Parqet Connect. It implements RFC 9728 (OAuth Protected Resource Metadata):

1. Parqet Connect (connect.parqet.com) = OAuth authorization server that issues access tokens and enforces the granted permissions
2. MCP Server (mcp.parqet.com) = protected resource that accepts bearer tokens and uses them for Parqet Connect API requests
3. ChatGPT / Claude / other MCP clients = obtain and use tokens

When you connect through an MCP client, you'll be redirected to Parqet to authorize access. The MCP server then uses your access token to read or write data through the Parqet Connect API on your behalf.

Discovery endpoints:

Supported scopes:

Available tools

All tools require authentication. Read tools use readOnlyHint: true; write tools use readOnlyHint: false and idempotentHint: false. Write tools require portfolio:write and should only be called after you confirm the final inputs. Call parqet_get_user first when the portfolio IDs are unknown.

parqet_help

Returns the MCP's current tool-routing instructions and parameter conventions. Use it when the agent is unsure which Parqet tool to call.

Parameters: none

parqet_get_user

Returns the authenticated user's info and their accessible portfolios. Use the portfolio IDs from the response for all other tools.

Parameters: none

Returns: userId, state, permissions[] (each with resourceId and access level)

parqet_list_portfolios

Lists all portfolios with metadata (name, currency, creation date, connected brokers) in a single call.

Parameters: none

Returns: items[] with id, name, currency, createdAt, distinctBrokers

parqet_create_portfolio

Creates a portfolio after you confirm its final name.

Parameters:

Returns: the new portfolio's id, name, and direct Parqet URL

parqet_create_holding

Creates a non-security holding in an existing portfolio. Supported holding types are cash, custom, real_estate, insurance, p2p, and commodity. Securities and cryptocurrencies are resolved from an ISIN or symbol when activities are added, so they do not need to be created first.

Parameters:

All holding types accept an optional externalId. Cash holdings can specify currency and referenceAccountFor. Custom holdings require assetProduct and can include initial valuation quotes. Commodity holdings require the metal identifier plus amount, purity, and unit.

Returns: the new holding's id, portfolioId, type, and name. Use the holding ID when adding activities for it.

parqet_get_portfolio

Gets metadata for a single portfolio by ID.

Parameters:

Returns: id, name, currency, createdAt

parqet_query_portfolio

Returns a compact, purpose-built portfolio view. This is the preferred tool for most portfolio questions.

Parameters:

Returns: compact data for the selected view, including the interval and value currency

parqet_get_activity_summary

Summarizes activities by type, asset type, and currency and automatically follows pagination up to a configurable limit. This is the preferred tool for transaction totals and recent-activity questions.

Parameters:

Returns: grouped totals, counts, recent activities, and completion/pagination metadata

parqet_get_activities

Retrieves raw transaction rows for a portfolio. Results are paginated. Prefer parqet_get_activity_summary unless row-level details are needed.

Parameters:

Returns: activities[] (each with type, nested asset object discriminated by assetIdentifierType, holdingId, holdingAssetType, shares, price, amount, amountNet, currency, datetime, tax, fee, broker; sell and transfer_out also include realizedGains, realizedGainsNet, avgHoldingPeriod, buyAmountNet), cursor

parqet_add_activities

Adds a batch of 1-100 activities to an existing portfolio after you confirm the portfolio and final activity details.

Parameters:

Each activity requires type, currency, datetime, an asset identifier, and either shares plus price or amount:

Optional fields include tax, fee, description, broker, and externalId. Insurance and P2P activities do not support interest.

Returns: the overall status and requested, created, and failed counts, plus createdActivities and notCreatedActivities

parqet_get_performance

Returns the raw, nested portfolio performance payload. Prefer parqet_query_portfolio unless the complete holdings, quote, performance, or chart payload is needed.

Parameters:

Returns:

• holdings[] — current positions with asset (name, ISIN/symbol, type), position (shares, purchasePrice, purchaseValue, currentPrice, currentValue, isSold), per-holding performance (kpis, gains, dividends, fees, taxes, valuation), quote and startQuote (price, currency, exchange, fx), nickname, logo
• kpis.inInterval.xirr — annualized money-weighted return
• kpis.inInterval.ttwror — time-weighted return
• valuation.atIntervalEnd — current total portfolio value
• realizedGains.inInterval — profit from sold positions
• unrealizedGains.inInterval — paper profit on current holdings
• dividends.inInterval — dividend income
• fees.inInterval.fees — total fees paid
• taxes.inInterval.taxes — total taxes paid
• netAllocations.net — net total across positive and negative holdings
• netAllocations.base — larger absolute side used as the allocation base
• netAllocations.positive — positive bucket with total, shareOfBase, and a holdings map by holding ID
• netAllocations.negative — negative bucket with total, shareOfBase, and a holdings map by holding ID

Connect to the Parqet MCP server

Any MCP client that supports Streamable HTTP transport and OAuth 2.0 can connect:

{
  "parqet": {
    "url": "https://mcp.parqet.com"
  }
}

The server uses OAuth for authentication. Your client will be redirected to Parqet Connect to authorize access.

See Connect the Parqet MCP to Claude for a full walkthrough.

Usage examples

Example 1: Check portfolio value and top holdings

Prompt: "What's my portfolio worth and what are my biggest positions?"

The AI will call parqet_get_user to find your portfolios, then parqet_query_portfolio with the overview view to get the current valuation and largest holdings. You'll get a response like:

Your portfolio "Main Account" is currently worth €47,230. Your top 3 positions are:

1. iShares MSCI World ETF — €18,500 (39.2%)
2. Apple Inc. — €8,200 (17.4%)
3. NVIDIA Corp. — €5,100 (10.8%)

Example 2: Review dividend income

Prompt: "How much dividend income did I receive, and from which holdings?"

The AI calls parqet_query_portfolio with the dividends view. You'll see a breakdown of gross/net dividend income, taxes withheld, and which assets paid dividends.

Example 3: Analyze recent transactions

Prompt: "Show me all my stock purchases in the last month."

The AI calls parqet_get_activities with activityType: ["buy"], follows pagination, and filters the rows by date. You'll get a list of recent buys with asset names, share counts, prices, and total amounts.

Example 4: Compare portfolio performance

Prompt: "What's my annualized return across all portfolios?"

The AI calls parqet_get_user to discover all portfolio IDs, then parqet_query_portfolio with all IDs and the overview view. You'll see your XIRR (money-weighted return) and TTWROR (time-weighted return) across all accounts.

Example 5: Look up a specific holding

Prompt: "How many Tesla shares do I have and what's my profit on them?"

The AI calls parqet_query_portfolio with the holdings view and finds Tesla in the result. You'll see your share count, current value, purchase value, and unrealized gain/loss.

Example 6: Understand net exposure

Prompt: "What's my net exposure, and which holdings contribute to negative allocations?"

The AI calls parqet_query_portfolio with the allocation view. You'll get the overall net exposure plus a breakdown of the holdings that make up the positive and negative sides of the portfolio.

Example 7: Create a portfolio

Prompt: "Create a portfolio named Long-term investments."

The AI confirms the final portfolio name with you, then calls parqet_create_portfolio. The response includes a direct link to open the new portfolio in Parqet.

Example 8: Add an activity

Prompt: "Add my purchase of 10 Apple shares at $225 per share on September 15, 2026 to my Main Account portfolio."

The AI resolves the portfolio and asset, asks you to confirm the final transaction details, then calls parqet_add_activities. The result reports whether the activity was created or why it failed.

Write safety

• Write tools should only run after explicit user intent and confirmation of the final inputs
• parqet_create_holding must be used before adding activities for a new cash, custom, real estate, insurance, P2P, or commodity holding
• Activity externalId values are optional but strongly recommended as deduplication keys; they must be unique within a batch
• Activity batches can partially succeed, so clients must inspect createdActivities and notCreatedActivities before retrying
• A timeout or server error can leave the write outcome unknown; verify the current portfolio data before retrying
• Historical shares and prices should be entered exactly as traded, without adjusting them for later stock splits

Privacy & data handling

• The MCP server only accesses data you explicitly authorize via OAuth
• All communication uses HTTPS/TLS
• No conversation data is collected or stored by the MCP server
• Access tokens are scoped to the permissions you grant during OAuth
• You can revoke access at any time through your Parqet account settings

For details, see the Parqet Privacy Policy.

Support

If you run into issues with the Parqet MCP server:

• Email: [email protected]
• Developer Hub: developer.parqet.com