How to connect CB Insights MCP with Codex

Add a column to a watchlist's table, or reconfigure one already there. Every add reports one entry per column under `added`, each with a `status` saying what happened to it. Pass a custom column's reported `name` to set_watchlist_cell_values to fill in values — that name is always one this watchlist resolves, so use it as reported rather than the name you asked for: where the watchlist already had a column of that name, it comes back as `Name (#7)`, which is the form every column tool accepts. **This is also how you edit an existing column's options or format.** Pass the column's existing `name` and matching `type`, and it is edited in place rather than duplicated, reported as `status: updated`: - `type='select'` + `options` — adds those labels to the ones it already offers (pass only the new ones). `added_options` lists the new labels and `options` the resulting full set, each carrying the `id` a cell stores and the chip's `color`, as `get_watchlist_contents` reports them. Pass a `label` back to `set_watchlist_cell_values`, never an `id`. - `type='number'` + `number_format` and/or `decimal_places` — changes how its values are displayed, e.g. currency to percentage. - `type='date'` + `date_format` — changes how its dates are displayed. None of these touch stored values: added options leave existing ones (and their cells) alone, and a format change is display-only. Never delete and recreate a column to change its options or format — that wipes every value in it. A column's resulting settings come back on a create too, since unspecified ones take the type's defaults. `type` has to match the existing column for the edit to land. If this watchlist has a select/number/date column of that `name` and you pass a different one of those types, the call is refused rather than adding a second column under the same name; read the column's real type from get_watchlist_contents (preset='table') and pass that. A 'cbi' column is an existing CB Insights data column (Total Funding, Mosaic, Sector, HQ Country) or an audit column (Added date, Added by). Call get_watchlist_column_catalog first for its `id_column_type`. Its values are filled in by CB Insights and cannot be set here, so its entry carries no `column_id`. Adding one that is already on the watchlist is a no-op reported as `status: already_present`; re-adding a deleted audit column restores its previous values and reports `status: restored`. Several 'cbi' columns may be added in one call. Each outcome is reported under `added`, and an id the watchlist doesn't offer is listed under `errors` — the rest are still added. The watchlist must already contain at least one company. Type-specific arguments must match `type` — passing `options` for a text column, say, is an error rather than being ignored.

Add organizations to an existing watchlist (by id_list or name). Orgs already in the watchlist are effectively a no-op, and still come back in `added`. Unresolved inputs (e.g. ambiguous names) come back in `errors` and don't fail the rest. Resolved orgs that ListService's list-size cap silently dropped come back in `not_added` instead of `added`, with a `not_added_note` explaining it.

Create a new custom watchlist for the current user, optionally seeded with organizations. The new watchlist is automatically pinned for the user. Returns the new `id_list` and the orgs added; unresolved seed inputs (e.g. ambiguous names) come back in `errors` without failing the creation. Resolved seed orgs that ListService reported success for but didn't create come back in `not_added` instead of `added`, with a `not_added_note` explaining it.

For each of the requested companies or investors, returns acquisitions, M&A deals, mergers, buyouts, and takeovers, ordered by recency. Use `search_deals` for M&A deal filtering capabilities, including 50+ attribute filters.

For each of the requested companies, returns its CB Insights analyst rating (optimistic, neutral, or cautious) — the verdict CBI's analysts publish on a company's outlook — plus the analyst's written rationale. Analyst ratings cover a curated subset of companies, so many companies have no rating; those come back with an `error` saying so rather than an empty payload. Fields returned: - analyst_rating: the company's current rating. Every field below is omitted when the analysts have not set it, so a sparse object is expected. - rating: the verdict — optimistic, neutral, or cautious. - status: how live the verdict is — live (actively maintained), under_review (being revisited), coming_soon (rating in progress, no verdict yet), or no_longer_rating (coverage dropped; the verdict is historical). Treat anything other than live as stale. - ts_rating: when the rating was last set. - description: a short summary blurb about the rating. - note: the analyst's written rationale, with title, thesis (the core argument), signals (a list of {name, content} evidence points backing the thesis), and what_analysts_are_watching_for (what would change their mind). - rating_history: every recorded change to the rating, newest first. Only present when preset='full'; an empty list means no changes were recorded. Each entry has change_type (initiated, reaffirmed, upgraded, or downgraded), rating (the verdict as of that change), and ts_rating (when the change was made). Guidance: The rating and its note are written by human CB Insights analysts, not generated. - ALWAYS state the rating and attribute it to the analysts: "CB Insights analysts rate X optimistic", "the CB Insights analyst thesis is...", "CB Insights analysts are watching for...". NEVER restate any of it as your own view ("my thesis", "I'm cautious on", "what I'm watching for"). - Quote the note verbatim inside quotation marks: the thesis as one unbroken quotation, never fragments stitched together with your own connecting words, and each signal's content quoted too - a bold label followed by unquoted text is not enough. Do not paraphrase, summarize, or restyle it, and do not close with your own restatement of what you just quoted. - NEVER infer a rating, thesis, or rating change from other data (Mosaic score, funding, commercial maturity, sentiment, third-party analysts). When a company comes back with the no-coverage `error`, say CB Insights has no published rating for it rather than substituting your own assessment. - The full note is title + thesis + signals + what_analysts_are_watching_for together. Quote all of it for a direct question about the rating, the company's outlook, its investment / partnership / acquisition merit, or its competitive positioning. For a general company overview, quote the thesis only. For a broad analytical question (the risks and opportunities for X), the note alone is not the answer - build it from the wider data and use verbatim extracts from the note as supporting evidence. - When you quote only an extract, point the user to the rest of it and link this result's `citation.url`. Word that pointer as part of the sentence you are already writing rather than a boilerplate tagline. NEVER offer the full note when you have already quoted it in full - there is nothing left to send them to.

Return buyer/customer interview transcripts about a company — interviews with software buyers describing their evaluation, purchase, and experience with the company's products. Use when the user asks for qualitative buyer feedback, customer interviews, voice-of-customer content, or firsthand accounts of a company's products. Returns up to `limit` transcripts (default 20). Pass `limit=null` to return every available transcript — useful when the user asks for "all" buyer interviews or wants comprehensive voice-of-customer coverage. Fields returned: - transcripts: list of buyer interview transcripts. Each transcript has: - transcript_id: stable identifier for the transcript. - title: title of the interview transcript. - event_date: date the interview took place (ISO format, may be empty). - content: full body of the interview transcript.

For each of the requested companies, returns its cap table history: per-round share class terms sourced from state incorporation filings (certificate of incorporation), sorted by round. Includes shares authorized, par value, issuance price, liquidation preference and amount, conversion price, dividend terms, participation rights, anti-dilution provisions, voting rights, and percent owned when available. Primarily covers private US companies; many companies have no cap table data.

For each of the requested companies, returns its competitors, similar companies, peers, challengers, and incumbents. To filter competitors by product type and 50+ attributes, use `search_companies` tool.

For each of the requested companies or investors, returns its exits (e.g. IPOs, mergers, acquisitions, going public, dissolutions, SPACs, buyouts, and other liquidity events). To search exits across companies with 50+ filters, use the `search_companies` tool.

For each of the requested companies or investors, returns its funding rounds (e.g. pre-seed, seed, Series A-F, growth, late-stage), including round dates, amounts, lead and participating investors, and valuation ranges when available. For multi-company searches with 50+ attribute filters, use `search_deals` tool.

Predict when companies will raise their next funding round (funding window, fundraise timing, likelihood to raise, next round prediction). Returns the predicted fundraising window (start and end date), the company's current phase relative to that window (pre/in/post), the peer cohort of similar companies used to derive the prediction, the percentage of peer companies that historically raised a next round, and the median days to next funding. Use preset='full' to also include the time-to-funding histogram.

Return structured quantitative metrics about a company's employee count and open job postings. Use when the user asks for specific numbers such as headcount, headcount growth rates, number of open positions, or momentum score. Do NOT use for questions about hiring strategy, workforce strategy, talent philosophy, workforce trends, or hiring outlook — use `get_company_hiring_insights` for those. Fields returned: - total_headcount, total_headcount_as_of: most recent total employee count and the date it was measured. - six_month_growth_pct, one_year_growth_pct, two_year_growth_pct: headcount growth over the trailing 6 months / 1 year / 2 years, as a percentage. - headcount_history: monthly total headcount over the past 2 years (oldest to newest). Each point has as_of_date, headcount, and percent_change_pct (change from the prior data point, when available). - department_distribution, country_distribution: object with as_of_date and a points list. Each point has name, headcount, and pct_of_total (the headcount as a percentage of the sum of all categorized headcounts in that distribution; sums to ~100). Sorted by headcount descending. - open_positions: count of open job postings for the most recent target month. - target_month: month that hiring stats (open positions, momentum) were captured. - hiring_intensity: percentage of open job positions to total headcount, indicating the company's hiring rate relative to its current workforce size. - momentum_score: a CBI proprietary index used to compare hiring activity across companies, accounting for company size. A larger company hiring at the same rate of job openings as a percentage of total headcount will have a higher momentum score than a smaller company with the same rate. The average momentum score is around 5 out of 100. - momentum_score_percentile: the percentile ranking of a company's momentum score compared to all other companies in the dataset, indicating what percentage of companies have lower momentum scores. For example, a company in the 99th percentile means 99% of other companies have lower momentum scores, placing it in the top 1% of hiring momentum.

Return AI-generated narratives about companies' hiring activity, talent strategy, workforce trends, and hiring outlook, based on CB Insights analysis.

For each of the requested companies or investors, returns its investments (rounds it has backed, led, or co-invested in), including the portfolio company, round type, date, and amount. To filter investments across investors with 50+ attribute filters, use the `search_deals` tool.

For each of the requested companies, returns the investors (e.g. VCs, angels, lead investors, participating investors, backers) who have funded it, including round participation details and lead investor status. To filter investors by type, geography, fund size, or 50+ other attributes, use `search_investors` instead.

For each of the requested companies, returns the market maps (e.g. industry landscapes, sectors, ESP categories, verticals, segments) that include it, with the market name, description, and associated industries.

For each of the requested companies or investors, returns recent news articles (e.g. press releases, announcements, headlines, media coverage, mentions), including dates, titles, content snippets, source URLs, and other organizations mentioned in the same articles.

For each of the requested companies, returns forward-looking outlook data: Mosaic scores (with insights, percentiles, and 1-year deltas), exit probability (IPO/M&A) with contributing signals, and commercial maturity / technology readiness levels (TRL). Mosaic is a 0-1000 health/growth score for private companies (>600 signals a higher likelihood of IPO, unicorn status, or favorable exit).

Return profile overviews for companies or investors: name, description, website, HQ location, founded year, status (e.g. alive, acquired, IPO, dead), total funding, Mosaic score, headcount, sector/industry classification, market maps, the CB Insights analyst rating, and a list of key people (leadership) when available. Use this to look up a single company/investor, or to fetch many at once — e.g. to populate a table or compare a set of companies side by side. To find or filter companies and investors by 50+ attributes (rather than look up ones you can already name or identify), use `search_companies` or `search_investors`. Guidance: - profile.analyst_rating is written by human CB Insights analysts. Attribute it to them, never as your own view; quote its text verbatim rather than paraphrasing; and when you surface only part of the note, point to the rest at `citation.url` in your own words. Call `get_company_analyst_ratings` for the rating change history. Guidance: - profile.key_people lists up to 10 of this company's key people (leadership). Call `get_company_key_people` directly for more, or to control how many are returned.

For each of the requested companies, returns its business relationships (e.g. partnerships, suppliers, customers, vendors, licensees), ordered by recency. To filter by relationship type (i.e. client of company, integration partners), industry, or 50+ attributes to find connections between specific companies, use `search_partnerships` instead.

For each of the requested companies, returns revenue data, including estimated and reported revenue figures, revenue ranges, and historical revenue when available.

Return the full transcript for a single earnings call (quarterly results, analyst call, investor call) for a public company, including all speaker paragraphs in order. Defaults to the most recent earnings call. To compare multiple earnings periods, call this tool in parallel once per period.

Returns an overview of a technology market or industry sector: market name, description, and key features; the list of companies / competitors in the market ranked by CBI Mosaic score (company health/momentum), each with CBI's scorecard analysis (competitive analysis, customer outcomes, etc.); and market trends including industry and sub-industry classification, equity funding totals and deal counts over 1-year and 2-year windows, average Mosaic score, commercial maturity stage, average employee headcount and headcount growth, a multi-year equity-funding-and-deals time series, and recent exit activity (IPOs / acquisitions). Use this to research a market's size, funding activity, competitive landscape, top startups, and growth trends. Accepts either a known market id or a market name (resolved against the CBI market taxonomy).

Report how many CB Insights MCP tool calls the current user has made and how many their plan allows. Use this when asked about usage, remaining tool calls, quotas, limits, or after a call is rejected for exceeding an allowance. Returns a daily per-user figure and an annual figure for the user's team (or, for an account with no team, an annual figure for the account itself), each with the number of calls used, the limit, and the window it applies to. Keywords: usage, quota, limit, allowance, remaining calls, rate limit, how many calls, over limit, exceeded.

The CB Insights data columns that can be added to a watchlist. Use this to find the `id_column_type` to pass to add_watchlist_column with type='cbi'. Columns are grouped by category, and a column is identified by its id, not its name. `already_added` marks the columns the watchlist already shows. Adding one again is a safe no-op rather than an error. A search that matches nothing returns no categories and says so in `note`. Covers CB Insights data columns and the audit columns (Added date, Added by). Custom columns you fill in yourself are not listed here; create those with add_watchlist_column and type 'text', 'select', 'number' or 'date'. Readable for any watchlist, but add_watchlist_column refuses a watchlist with no companies in it.

Fetch the organizations in a specific watchlist, by id_list or name. The default 'table' preset returns the watchlist as the user configured it: `columns` in their order — CB Insights data columns (Mosaic, Total Funding, headcount) alongside their own Notes and Funnel columns — and a `values` map per row keyed by column name, each cell a list of the values it holds. A value has `text`, plus an `id`, `key`, `url`, `logo` or `subtext` where it has one. This is what the watchlist shows in the app. A select/funnel column's header carries `options`, the full set that column offers, each with its `id`, `label` and chip `color`. The `label` is what `set_watchlist_cell_values` accepts for it, and the set is not the same as the subset the visible rows happen to use. A number column's header carries `number_format` and `decimal_places`, so how it's configured is visible without inferring it from the values on the page. Pass `sort_field` to order the whole watchlist by a column rather than the page, taking the id off a column header in an earlier response; a manual or AI column needs its `matrix_column_id` too, since those families share one `id_column_type`. `is_sortable` on the header says whether a column can order at all. The 'orgs' preset returns lightweight stubs instead (id_org, name, url, status, latest funding), for chaining into the `get_company_*` tools. Only approved members are returned — the same set the user sees when they open the watchlist in the app. Both removed (rejected/deleted) entries and pending suggestions (not yet approved or rejected) are excluded. Rendering: the client displays this response as an interactive table the user can page through and export. They can already see the rows, so do not list them back — reply with at most one line of context, or just the answer to what was asked. A request to see, show or open a watchlist needs exactly one call; the table fetches its own further pages. Pass `offset` only to answer a question that genuinely needs rows beyond the first page. "Full", "whole" or "all" in such a request means every *column*, which the default 'table' preset already returns — not every row. The 'table' preset returns a fixed 10-row page, so a larger watchlist is never returned whole. Do not page to the end to assemble the rows, and never write them to a file or build a CSV: the table has its own CSV export and a link to the full watchlist, so point the user at those instead. To rank or pick across the whole watchlist, pass `sort_field` rather than paging — it orders every member, so the answer is on the first page.

Return the current user's watchlists, selected by a ListSet. Each watchlist comes back as {id_list, name, type (custom/competitors), item_count, is_pinned, permission_level, can_rename, watchlist_url}, where watchlist_url links to the watchlist in the CB Insights app. With no selector (or an empty one) this lists all of your watchlists; with containing_orgs it finds which of your watchlists contain those orgs. Use a returned id_list with `get_watchlist_contents` to see its members, or with the `get_company_*` tools' `selector.watchlist_ids`. `item_count` counts approved orgs, pending suggestions (not yet approved or rejected), and non-org rows such as people — so it can be larger than `get_watchlist_contents`'s `total` for the same watchlist, which counts approved orgs only: the same set the user sees when they open the watchlist in the app. Removed orgs count toward neither. `permission_level` is your access to the list — owner for your own lists, read_write or read for one shared with you. `can_rename` tells you whether `rename_watchlist` will accept it: false for a read-only share and for the auto-generated competitors list. The two `can_rename: false` cases differ for membership edits, though: the competitors list still accepts `add_watchlist_orgs` / `remove_watchlist_orgs`, but a read-only share (`permission_level: "read"`) refuses those too — only `permission_level: "read_write"` or `"owner"` can edit a list's members. Unresolvable watchlist references (list_names, list_ids) or org references (containing_orgs) are reported in `errors` with near-matches where available and don't fail the rest. An id_list you don't own comes back as an error, not as an empty result — so an empty `watchlists` with no `errors` genuinely means the selection matched nothing.

Fetch a page of results from an existing company search. Use this after the initial search to page through results beyond the initial limit, or to re-sort without re-running the search. Pass the search_id returned by the initial search along with limit and offset. Only pass a sort_field to re-sort; omit it and pagination keeps the search's default (relevance) order.

Rename a watchlist (targeted by id_list or its current name). Requires that you own the watchlist or hold write permission on it. The auto-generated competitors watchlist can't be renamed.

Rename a custom column on a watchlist, or one of its funnel options. **This is how you rename an option in a select/funnel column.** Pass the column in `column`, the option's current label in `option`, and its new label in `new_name`. The option keeps its identity, so every org that had it selected keeps that value and simply shows the new label. Do NOT rename an option by removing it and adding a replacement — that looks equivalent and isn't. Removing an option destroys the cells that used it, and the replacement is created fresh, so those values do not come back. `renamed` reports whether the column or an option changed; for an option, `previous_option`, `option` and the column's resulting `options` — in the shape `get_watchlist_contents` reports them — come back too. Without `option`, the column itself is renamed (only columns you added).

Find a population of companies matching criteria described in natural language. Industry, market, and technology concepts are matched as keyterms; the attributes below become structured filters. Returns filtered (not ranked) results, so ranking asks such as "top 10 by revenue" come back unsorted — sort them yourself if order matters. Use it for discovery: "AI drug discovery startups in Boston that raised a Series B", "companies acquired by Walmart in logistics", "B2B SaaS companies with 20%+ headcount growth". Do NOT use it to look up or verify one specific named company, or to check attributes of a list of named companies ("Klarna", "AWS ProServe", "company named Evergrow", "are X, Y and Z based in India") — those are rejected; use `get_company_profile` or another `get_company_*` tool. Naming a company as a criterion is fine ("backed by Sequoia", "acquired by Walmart"). For competitors of a named company prefer `get_company_competitors`; company search can be used but is weaker for large multi-segment companies (Apple, ExxonMobil). Available filter attributes (mention any of these in `query` and they are applied automatically, as inclusions or exclusions): geography (headquarters), business model (B2B, B2C, B2G, SaaS, marketplace, freemium, usage-based, etc. — only when the query names one explicitly), company status, commercial maturity, headcount, headcount growth 6m, headcount growth 12m, headcount growth 24m, revenue, revenue growth latest, revenue growth future, revenue multiple, revenue per headcount, stock price, market cap, mosaic overall, mosaic momentum, mosaic money, mosaic market, mosaic management, M&A probability, IPO probability, founded year, total funding, latest funding amount, latest valuation, deal size, exit valuation, historical investment stages, funding round, funding date, deal date, exit date, funding window, VC backed, analyst rating, exit type, investor, lead investor, acquirer, company name. Guidance: REPORTING RESULTS TO THE USER — read before writing any summary: The rendered result view already gives the user full, independent access to every matched company (all `total_results` of them), with their own pagination — this is not limited by how many rows you retrieved into this conversation. Never describe your own retrieval as if it were the user's view. Do NOT say "showing the first 50" or "pulled 50 companies" — the user isn't limited to 50 of anything; they can browse the entire matched set themselves right now. - State `total_results` as the headline number: "matched 5,488 companies" — full stop, that's what the user has. - If you personally read some rows into your own context to reason, summarize, or answer a follow-up question, describe that as your own review, separate from what the user can see — e.g. "I looked at a sample of these to summarize common themes" or "I reviewed 150 of the 2,971 matches to answer your ranking question." Never phrase your own retrieval count as if it were the size or limit of the result set itself. - If total_results is large and the user's request implies judgment/ranking over the whole set (e.g. "who's the best X", "find the top candidate for Y", "which of these is most Z") rather than a structured filter: don't answer from just your first tool call's rows. Page through more of the result set in batches using `paginate_and_sort_search`, scoring/filtering each batch against the user's rubric yourself, keeping only the survivors. State explicitly how many rows you personally evaluated (e.g. "evaluated 300 of 2,971 matches") — as a description of your own analysis depth, never as a description of what the user has access to. Caveats: - Results may include some off-target companies — that is the nature of search over keyterm matching. Review the returned companies and drop rows that don't fit the intent before presenting them. Or run post processing logic to further filter or validate the list against additional custom filters. - Geography is headquarters location only. There is no filter for operations, presence, expansion into, or adoption of something in a place, so such criteria go unsearched. - Highly specific criteria — niche sub-sectors, several stacked quantitative constraints, obscure investor portfolios — can return very few or zero companies. When results are thin, drop or loosen the narrowest criteria, or split the query, instead of stacking more. - Criteria are AND-combined. Genuinely alternative conditions ("funded OR acquired by X", "raised in the last 90 days OR grew headcount 20%+") need one search each. - Prefer several narrow searches over one sprawling multi-sector search. Always show the citation source URL in your answer - the results must be cited.

Find funding deals (financing rounds and investments) matching criteria described in natural language. Industry, market, and technology concepts are matched as keyterms; the attributes below become structured filters. Each result is a deal: the company that raised, the round, amount, date, and participating investors. Results are filtered, not ranked. Use `search_companies` to get companies rather than individual rounds, `search_investors` for a list of firms, and `search_partnerships` for non-financing deals (licensing, partnerships, supply agreements). `search_companies` is also more effective in seeing which companies a particular company invested in or acquired. Available filter attributes (mention any of these in `query` and they are applied automatically): geography, deal size, deal date, investment stage (funding round), investor, lead investor. Caveats: - Geography is the raising company's headquarters; there is no filter for where it operates. - Anything outside the attribute list above (deal rationale, use of proceeds, valuation multiples) cannot be filtered on and goes unsearched. - Highly specific criteria (a niche sub-sector plus a narrow date window plus a specific investor) can return very few or zero deals. When results are thin, loosen or drop the narrowest criteria rather than adding more. Always show the citation source URL in your answer — the results must be cited.

Find a population of investors — venture capital firms (VCs), private equity funds (PE), angel investors, family offices, corporate venture arms, hedge funds — matching criteria described in natural language. Portfolio-company industry, market, and technology concepts are matched as keyterms; the attributes below become structured filters. Each result is an investor with their type, location, and investment activity. Results are filtered, not ranked. Use `search_companies` to find the portfolio companies themselves, `search_deals` for individual financing rounds, and `get_company_investors` / `get_company_profile` to look up one named firm rather than discovering a set. A query that is just a firm's name, or a list of named firms to check facts against, is rejected — name investors only as criteria ("co-investors with Sequoia"). Available filter attributes (mention any of these in `query` and they are applied automatically): investor name, investor geography, investor type, investor founded year, investment stage, total funding, deal size, valuation, deal date, co-investor, deals last 12 months, total deals, exit valuation, exits last 12 months, portfolio company name, portfolio company status, portfolio company founded year, portfolio company location. Caveats: - Geography is headquarters location only; there is no filter for where a firm operates or is expanding. - Highly specific criteria (narrow sub-sectors plus several stacked constraints) can return very few or zero investors. When results are thin, loosen or drop the narrowest criteria. Always show the citation source URL in your answer — the results must be cited.

Find existing business relationships between companies — partnerships, alliances, collaborations, vendor/client arrangements, licensing, supply/distribution, sponsorships — matching criteria described in natural language. Results are filtered, not ranked. Use it for: partnerships between two industries, cross-sector collaborations, vendor relationships in a market, licensing deals, supply chain connections, sponsorships, strategic alliances. Named companies are expected here as the partner sides. Use `get_company_relationships` for every business connection of one named company without criteria, `search_deals` for financing rounds, and `search_companies` for hypothetical partners ("companies that could partner with X") — this tool only returns relationships that already exist. Relationship-level filters: relationship type (partnership, vendor/client, licensor/licensee, supplier/distributor, sponsorship), relationship keywords, time period, and one or two specific companies or described company populations as the partner sides. When a partner side is described rather than named, it supports the full company filter set (mention any of these in `query` and they are applied automatically): geography (headquarters), business model (B2B, B2C, B2G, SaaS, marketplace, etc. — only when named explicitly), company status, commercial maturity, headcount, headcount growth 6m, headcount growth 12m, headcount growth 24m, revenue, revenue growth latest, revenue growth future, revenue multiple, revenue per headcount, stock price, market cap, mosaic overall, mosaic momentum, mosaic money, mosaic market, mosaic management, M&A probability, IPO probability, founded year, total funding, latest funding amount, latest valuation, deal size, exit valuation, historical investment stages, funding round, funding date, deal date, exit date, funding window, VC backed, analyst briefing, exit type, investor, lead investor, acquirer, company name. Caveats: - Geography is headquarters location only; there is no filter for where a company operates. - Relationship coverage is sparser than company coverage: a specific relationship type plus a niche sector on both sides plus a short time window often returns very few or zero rows. When results are thin, widen the time period, drop the relationship type, or describe only one side. Always show the citation source URL in your answer — the results must be cited.

Find people (executives, founders, engineers, etc.) by their role and background. Filters are extracted automatically from the natural-language query. Results are filtered, not ranked. Use it when the answer should be a list of PEOPLE ("Heads of Marketing at AI startups", "ex-Salesforce executives now at seed-stage companies"). Use `search_companies` when the answer should be COMPANIES, even if selected by their team ("which AI companies have founders from Salesforce"). The search supports four optional, AND-combined sections: - Person Details: name, location (the person's location), contact availability (email / LinkedIn), education degree, and leadership signals (below). - Current Role: job title, seniority level, department, time in current role, plus the company/industry the role is at. - Past Role: the same job attributes (no time-in-role) for a role the person previously held. - Current or Past Role: a role the person holds OR has held. Available leadership signals (career-spanning achievements; mention any in `query`): repeat founder, technical founder, Y Combinator founder, gave an executive interview to CB Insights, top university alum, founder money raised, valuation achieved, revenue achieved, has IPO'd, has M&A'd, has acquired companies, investment stage experience (Series A, B, C, etc.), and early employee at a top private or recently public company. The company/industry side of any role reuses the full company filter set (industry, market, technology, geography, headcount, revenue, funding, status, etc.) — describe the company in natural language. Available company filter attributes (mention any of these in `query` and they are applied automatically): geography (headquarters), business model (B2B, B2C, B2G, SaaS, marketplace, etc. — only when named explicitly), company status, commercial maturity, headcount, headcount growth 6m, headcount growth 12m, headcount growth 24m, revenue, revenue growth latest, revenue growth future, revenue multiple, revenue per headcount, stock price, market cap, mosaic overall, mosaic momentum, mosaic money, mosaic market, mosaic management, M&A probability, IPO probability, founded year, total funding, latest funding amount, latest valuation, deal size, exit valuation, historical investment stages, funding round, funding date, deal date, exit date, funding window, VC backed, analyst briefing, exit type, investor, lead investor, acquirer, company name. Each result is a person with their CB Insights profile URL (``profile_url``), location, contact info, and roles (job title, seniority, tenure, plus a ``company`` object carrying the org's name, ``id_org``, website ``url``, ``logo_url``, and CBI ``profile_url``). Caveats: - Person location and company geography are headquarters/home locations, not where someone operates or is relocating to. - Stacking a narrow job title, a niche company population, and a tight tenure window often returns very few or zero people. When results are thin, loosen the title or drop the narrowest constraint instead of adding more. Always show the citation source URL in your answer — the results must be cited.

Find CB Insights proprietary research (analyst content) matching a topic and criteria described in natural language. Topics, industries, and keywords are matched as keyterms; the attributes below become structured filters. Results are filtered, not ranked. Use it for analyst research only — for company, deal, or investor data use `search_companies`, `search_deals`, or `search_investors` respectively. Available filter attributes (mention any of these in `query` and they are applied automatically): research topic / keywords, industry, specific organization, research type, publish date. Research types: market maps, research briefs, state of reports, future of reports, competitor analysis, top company lists, big tech reports, buyer perspective reports, investment thesis maps, customer sentiment, webinars, markets. Caveats: - Without an explicit date in the query, results are limited to the last two years. - Research coverage is a finite analyst library: a narrow topic combined with a specific research type, organization, and date window often returns very few or zero reports. When results are thin, broaden the topic or drop the research type rather than adding criteria. Always show the citation source URL in your answer — the results must be cited.

Set values in a watchlist's custom columns for one or more orgs. Unresolved orgs, orgs that aren't in the watchlist, unknown columns, values that don't fit a column's type, and two keys landing on the same cell are reported in `errors` (with alternates) and don't fail the rest. `set` lists only what was written — trust it over assuming a value survived, since a failure partway through a select column leaves that cell empty.