MCP

Overview

Primary Portal MCP lets you query Primary Portal’s ECM deal data directly from your own AI assistant (Claude, ChatGPT, or any other MCP-compatible tool) using natural language instead of the web UI. It exposes deal search, filtering, aggregated metrics, deal lookups, lock-up expiries and bookbuild messages over the Model Context Protocol, scoped to whatever your account and organisation are entitled to see.

Before you start

You’ll need:

• An existing Primary Portal investor account. This surface is investor-only.
• A Priced or Live entitlement on your organisation. Live includes everything Priced does, plus announced/live deals and bookbuild messages (see Entitlements). If you’re not sure which you hold, ask your Primary Portal account team.
• For announced or in-flight (not yet priced) deals from a specific bank, an active connection to that bank. Without one, you’ll only see the priced-deal universe your entitlement covers.

Connecting your AI assistant

Endpoint. Primary Portal MCP is reached at:

https://mcp.live.ai.primaryportal.com/mcp

Authentication. OAuth 2.1, with no API key or personal access token to generate or store. When you add Primary Portal MCP as a remote server in your AI assistant, it will redirect you to sign in with your existing Primary Portal investor credentials and ask you to authorise access.

Steps (exact menus vary by client):

Once connected, the tools your assistant can see are filtered live to your organisation’s entitlements, so what you’re offered may differ from a colleague’s, even at the same firm.

Entitlements

Access comes in two tiers, and Live is a superset of Priced:

Your entitlement decides both the deal universe you can query and which tools your assistant is offered. The catalogue is filtered live at connection time, so a Live user sees more tools than a Priced user, and what you’re offered may differ from a colleague’s, even at the same firm.

Available tools

Every tool Primary Portal MCP can expose is listed below. The Access column shows the minimum entitlement. A Priced tool is also available to Live users, since Live includes Priced.

All tools are read-only. Rather than treating this table as fixed, ask your assistant to list its available tools after connecting, or call get_deals_data_coverage.

Deal tools

Five tools cover the deals dataset. Most questions follow the same path:

1. get_available_deals_filters to find valid filter values.
2. search_deals to find deals, or get_aggregated_deals_metrics to summarise them.
3. get_deal for the full record of specific deals.

get_deals_data_coverage explains what each field means and how often it’s filled.

Filters

search_deals and get_aggregated_deals_metrics take the same filters. All are optional, and they combine with AND.

search_deals

Finds deals that match your filters, most recently priced first. Withdrawn deals are ordered by withdrawal date.

• Needs at least one filter. state=PRICED on its own doesn’t count, but ANNOUNCED, LIVE or WITHDRAWN does.
• Returns 20 deals per page by default and up to 50, with the total count, whether there are more, and the offset of the next page.
• Each deal carries every field in What a deal contains except state and marketCapBand. Use get_deal for those.

Examples: offeringType IPO, countries ["IND"] and pricingDateFrom 2026-01-01 lists IPOs listed in India and priced in 2026. state WITHDRAWN and withdrawnDateFrom 2026-01-01 lists deals withdrawn this year.

get_aggregated_deals_metrics

Groups deals by one dimension and returns one metric for each group.

• groupBy: sector, region, offeringType, marketCapBand, state, country or publishingBank.

• metric: one of the following.

  Metric	What it returns
  dealCount	Number of deals
  totalVolumeUSD	Total deal size, in US dollars
  avgPerf1d / avgPerf1w / avgPerf1m	Average share price move one day, one week or one month after pricing, in percentage points

• Filters: the same as search_deals. For dates, use one of the pricing-date range, lookbackDays (days back from the latest pricing date in the data) or the withdrawal-date range.

• States: without a state or date filter, groups include announced, live and withdrawn deals as well as priced ones. Set state PRICED, or a pricing-date range, to count completed deals only.

• Returns one row per group: the group value, its deal count and the metric. Check the deal count before trusting an average.

Example: groupBy country, metric totalVolumeUSD, offeringType IPO and lookbackDays 90 gives IPO volume by listing country over the last 90 days.

get_deal

Returns the full record for up to 50 deals at once.

• dealIds: an array of 1 to 50 deal IDs. IDs from search_deals, search_lockups and search_bookbuild_messages all work.
• Returns the deals in the order you asked for, every field in What a deal contains, plus a list of any IDs that matched no deal.

get_available_deals_filters

Returns the valid values for every filter above, worked out from the data, plus the earliest and latest pricing dates. Pass values to the filters exactly as returned. Bank names are a long list, so they’re only included when you pass includePublishingBanks=true.

get_deals_data_coverage

Describes the dataset: the pricing-date and withdrawal-date ranges, deal counts by state, and every field the deal tools return, with what it means, its possible values and how often it’s filled. It also lists the dataset’s limits. Ask your assistant to call it when you need to know how far an answer can be trusted.

Publishing banks

You can filter and group deals by the banks on them, in any role: bookrunners and coordinators, but also sponsors, co-managers, placing agents and advisers.

• Any bank name containing your term counts, so “Morgan Stanley” also matches “MUFG Morgan Stanley”, and a broad term such as “Bank” pulls in several banks.
• If a name matches no bank, you get no deals back, plus a list of similar bank names to try instead.
• Grouping by publishingBank uses the bank names as they appear on each deal, so one bank can show up under several name variants. Check the groups before comparing two banks.

Countries

countries filters on the country of the listing exchange, not the issuer’s home country. Values are three-letter ISO 3166 codes, for example IND, JPN or GBR. A dual-listed deal matches each of its countries.

What a deal contains

get_deal returns every field below. search_deals returns all of them except state and marketCapBand. Not every field is filled for every deal: announced and live deals have no pricing date, prices or performance yet, withdrawn deals never get them, and EQL deals have no prices, performance or market cap.

Sample prompts

• “What deal data do I have access to?”
• “Show me IPOs priced in the technology sector in EMEA over the last 6 months.”
• “What’s the average one-week aftermarket performance for healthcare IPOs this year, by region?”
• “Show me the deals Nomura was on in the last six months.”
• “Compare Goldman Sachs and Morgan Stanley on healthcare IPOs this year by deal count.”
• “List IPOs listed in India in 2026.”
• “Which country had the highest total ECM volume in APAC this year?”
• “Which IPOs were withdrawn in EMEA this year?”
• “Give me full details on these three deals: [deal IDs].”
• “What filters can I use to narrow a deal search?”
• “Which IPO lock-ups are expiring in the next month in EMEA?”
• “Which lock-ups expire in Japan in the next two weeks?”
• “Show me the bookbuild messages from JP Morgan for [issuer name].”

Lock-ups

search_lockups queries an upcoming lock-up expiries calendar: a forward-looking view of when share lock-ups on recent deals are due to expire. It’s available on the Priced entitlement, and therefore to Live users too.

Filters:

• dealTypes (required): one or more offering types.
• issuer (optional): partial and case-insensitive; matches on both the issuer name and the ticker.
• region (optional): a specific deal region.
• countries (optional): one or more listing countries (see Countries).
• expiryFrom / expiryTo (optional): bound the results to a lock-up expiry date range.
• Results are ordered by expiry date, nearest first, and paginated.

Each row also lists the deal’s bookrunners and a link to the deal in Primary Portal.

The dataset is a rolling snapshot of upcoming expiries (currently around a month ahead), not a full history. Depending on when it was last loaded, it can also include expiries from the past few days. Every response reports the expiry window it currently holds, so an empty result tells you whether nothing matched your filters or you simply queried outside the loaded window.

Bookbuild messages

If your organisation holds the Live entitlement, two additional tools let you search bookbuild messages.

• search_bookbuild_messages: filter by dealIds (exact match), issuers (partial, case-insensitive) or bankNames (partial, case-insensitive). At least one of these is required. Each takes up to 50 values: values within a filter are combined with OR, and different filters with AND. You can narrow further with countries, but it doesn’t count as the required filter on its own. Results are paginated (up to 50 per page), ordered most recent first, and message bodies are returned in full. Each message carries its deal’s state, including withdrawn, and a link to the deal.
• get_available_bookbuild_filters: takes no input and returns the issuer names, bank names and listing countries present in the messages you can see, so you can discover valid filter values instead of guessing (the bookbuild equivalent of get_available_deals_filters).

You only see messages for deals you can see through the deal tools. A deal ID you can’t see returns an empty list rather than an error, as does a search with no matches.

What data you’ll see

Announced and live deals. Priced users see the priced-deal universe. Live users additionally see announced and in-flight deals, but only where there’s an active connection to the relevant bank, and further filtered by country and regulatory-status rules (REG_S vs QIBS). If a connection is revoked, it takes effect on your very next query.

Deal states. Deals are PRICED, ANNOUNCED, LIVE or WITHDRAWN. Withdrawn deals were cancelled before pricing. They’re included on the Priced entitlement, and get_deals_data_coverage reports the span of withdrawal dates available.

Links to Primary Portal. Every deal, lock-up and bookbuild result includes a link to the deal in Primary Portal: the termsheet for announced and live deals, and the deal page otherwise. Your assistant should pass the link on unchanged rather than build its own.

Input rules.

• Fixed values such as region, offering type, market-cap band, state and deal type are accepted in any case, so ipo works as well as IPO.
• Filters that take a list (issuers, countries, banks, deal IDs) accept up to 50 values. More than that is rejected with an input error.

Fair use thresholds

Primary Portal MCP isn’t charged by volume, but use must stay within these thresholds, under clause 5.2 of the Data MCP Terms of Use.

Use above any threshold is presumed to be outside interactive sessions, unless you can show otherwise.

Security and data handling

• No API keys or personal access tokens: you sign in with your existing Primary Portal account (OAuth), and access ends as soon as your account, connection or entitlement changes.
• Every query is checked against your current entitlements and connections, so a change applies from your next query.
• You only ever see the deals your organisation is entitled to.
• Every tool is read-only: none can create, change or delete anything.
• Access is logged for usage and audit purposes.

Getting help