Connect your AI assistant to conclusia

One endpoint serves every client. Sign-in is OAuth with your conclusia login - no client id, no secret. Any other client that speaks remote MCP with OAuth works the same way: add the endpoint as a server and sign in when it asks. Asking What lists do I have in conclusia? should show your lists; being asked to sign in again after a while is normal, and signing in is all it takes.

• Claude. On a Team or Enterprise plan an owner enables conclusia once, and everyone in the workspace can use it - each with their own conclusia login. On an individual plan the steps are yours. You stay signed in for 30 days, and Claude renews the session on its own while you keep using it.
• ChatGPT. Reading is silent, but ChatGPT asks you to confirm before anything that changes a list, and a list build runs past the end of an answer: ChatGPT starts it, tells you it is running, and picks the result up when you ask again.
• Microsoft 365 Copilot. The one client without the endpoint: Copilot does not let a user paste one, so your Microsoft 365 administrator installs the conclusia agent once for the whole organization. Users then open Copilot Chat, pick conclusia under Agents and ask; the first question brings up a Sign in to conclusia prompt. Copilot cannot wait for a list to be built: it tells you the run has started and how long it usually takes; ask again a few minutes later and it reads the result.
• Copilot Studio. For your own agents, add conclusia as a tool instead of installing the agent package. Each user sees only their own lists, and because the connection is a Power Platform connector, your data policies for connectors apply to it.
• Langdock. conclusia is an integration: someone allowed to create integrations (or a workspace admin) adds it once, and everyone connects with their own conclusia login. In a Langdock agent, add the conclusia actions and let users connect with their own credentials; a preselected connection would make everyone work in one person's conclusia account. Langdock runs a tool without asking unless you turn confirmation on for it - turn it on for the tools that change a list: create_list, refine_list, set_list_filters, add_research_column and report_problem. On a dedicated Langdock deployment with its own domain, write to support@conclusia.io first, so we can allow its sign-in address.

You do not need a contract to try this. Create an account at app.conclusia.io/free-trial, then connect your assistant the same way a customer does - the endpoint and the login are the same.

The free plan builds a whole list and shows you the first 20 companies on it plus every company you imported yourself, 20 lists in total, 10 messages per list and 3 list updates per list. Research columns run against the research balance your workspace starts with.

Locked on the free plan: sharing a list, removing companies from it and client reports. Ask for one and conclusia says so in one sentence and points at the upgrade in the web app (app.conclusia.io/profile). Everything else on this page works.

Exporting works on the free plan, through your assistant and with the Excel button in the web app, with the companies the plan shows you: every company you imported plus the first 20. The file says on its last line how many more are on the list.

Fourteen, no more:

• list_cases — your lists.
• get_list — one list: its criteria, a summary, and a page of companies.
• create_list — a new list from a description of what you are looking for.
• import_list - a new list from companies you already have, for example a spreadsheet: no analyst run, ready for research columns at once.
• refine_list — a change to a list that already exists.
• exclude_companies - named companies off a list at once, by website or name; they stay off through every later change.
• set_list_filters — narrow or widen a list by country, region or company size.
• export_list — the whole list as a downloadable Excel file.
• plan_research_column — the cost of a new research column, before anything runs.
• add_research_column — a priced research column, now running.
• report_problem — tells us when conclusia could not do what you asked.
• company_lookup — what conclusia knows about one company, by website or name.
• find_markets: searches conclusia's market structure by free text.
• get_market: opens one market, or lists the top-level sectors.

find_markets and get_market are free to use with any conclusia account, including the free plan: they walk the same market structure conclusia's analyst classifies lists into, with a name, a description and an approximate size for each market, but never a company name. When you name a market or industry, your assistant is told to start here and show you what the tree actually holds before building anything, so you pick the market rather than your assistant guessing it. Pass a hit's market_id (or a few) to create_list's market_ids and conclusia's analyst starts from that market: it checks every company in it against your brief, so the brief still has to say what kind of company you want. It is a normal build that takes minutes, not a shortcut around the analyst. It requires an account like any list, shows the free plan's usual first twenty companies, and refuses markets totalling more than 10,000 companies in favor of opening the market with get_market and picking a child instead. refine_list changes the criteria afterwards, and set_list_filters narrows the list by country, region or size, without touching the criteria.

import_list turns companies you already have into a conclusia list without an analyst run. Your assistant passes the rows - a website or company name for each company, plus your own columns - up to 500 per call, and appends further parts to the same list. Companies conclusia does not know yet are crawled and fill in over the next minutes; names that match several companies are left out and named, so you can send their websites instead. A list holds as many imported companies as your plan allows per upload. For a large spreadsheet the file upload in the conclusia web app is quicker, because nothing has to be retyped. Research columns cover every imported company, on the free plan too, and are quoted before they start.

exclude_companies takes companies off a list the moment you name them, without an analyst run. Websites match exactly; a name has to match one company on the list, and a name that fits several comes back with their websites so you can pick. Removed companies stay off through every later change to the list. A requirement such as "no private-equity-owned companies" is not a list of names: your assistant passes that to refine_list. On the free plan the companies are downvoted instead, which hides them the same way.

Your assistant can still brief create_list in plain language instead, naming the market, when you want the analyst to research it rather than load it directly.

On the free plan export_list returns the companies the plan shows (every imported company plus the first 20), and the file names how many more the list holds. create_list, refine_list and set_list_filters work until one of the free plan's counts is used up, and then say so.

If your assistant runs into something conclusia cannot do for you, it can say so to our team with report_problem — the same place the feedback button in the web app goes. It will tell you what it reported. Nobody answers in the chat, so treat it as a note to us, not as a support ticket.

That report is the one place where text from your conversation leaves conclusia: the two sentences the assistant writes (what you wanted, and what did not work) land in our team's Slack, together with the list's name and your account id. Nothing else you say to your assistant is sent anywhere, and no other tool forwards conversation text. The assistant only ever calls it when you have asked for something conclusia could not do.

There is deliberately nothing here that deletes, archives or shares a list. Those stay in the web app, where you can see what you are doing.

company_lookup answers a question about one company without a list being involved: ask "what do you have on acme.de" and your assistant reads conclusia's base and tells you the country, region, size band, industry and a short summary.

Two things worth knowing. It reads what conclusia already holds — it does not go and look at the company's website for you, so the answer is as fresh as our last crawl, and a company conclusia has never seen simply is not found. And the website is the reliable way to ask: company names are not unique, so a name matching more than one company is refused rather than guessed.

There is a daily limit per account, high enough that normal use never meets it.

Ask for the list as a file and your assistant will call export_list. That returns a download link (valid for one hour) to an Excel file with every company and every research column — your assistant cannot open the file itself, so if you want it to work with the full list, download the file and drop it back into the chat.

On the free plan the file holds the companies the plan shows you (every company you imported plus the first 20), and ends with one line saying how many more are on the list. The About sheet repeats it. A paid plan exports the whole list.

A list can carry researched columns: one question answered for every company on it, gathered from the open web. "How many units do they manage", "who owns them", "do they sell to public bodies" — once a column exists, the list can be filtered and compared on something it did not contain before.

Your assistant creates one in two steps.

1. plan_research_column works out what the column would cost. It returns the number of companies, an estimated range, and your account's remaining research balance. It starts nothing and charges nothing.
2. add_research_column starts the research, using the plan from step 1. This spends your account's research budget.

The split is deliberate: research is paid per company, so the cost reaches you before the decision, not after it. Your assistant should show you the estimate and wait for your answer. A plan is good for fifteen minutes.

Answers arrive company by company over the following minutes. get_list shows progress under research_column_details and the answers themselves appear on each company under the column's name.

Pausing a running column, refilling one after new companies arrive, and deleting one are done in the conclusia web app, on the list itself.

A list can hold thousands of companies, so get_list returns them a page at a time - 25 by default, 100 at most. Your assistant pages through them by itself when it needs to; you can also just ask for "the next 50".

Each company carries:

• name, website, country, region, headcount band, industry, a short summary
• whether you upvoted or bookmarked it
• its market segment, when the list defines segments — the same labels the web app's legend shows
• the answers to that list's research columns — the questions you had conclusia research per company

A missing answer means that question has not been researched for that company yet, and null means conclusia looked and found nothing. Your assistant is told the difference, so it will not report one as the other.

That reading of null holds for open-text and structured columns. A multiple-choice column works differently, and it is worth knowing: it has to land on one of the options you gave it, so a search that turns up nothing is recorded as whichever option means "nothing found" rather than as null. On the two curated topics that option is Independent for ownership and No Activity for M&A — so a company reading Independent may mean "we checked for an investor, a listing and a parent company and established that it is independent", or it may mean "we looked and found nothing at all".

conclusia stores a short reason behind every answer, and that reason is what separates the two. Ask your assistant for the reasoning and it reads the list with include_reasons, which returns those reasons per company under research_reasons, next to the answers themselves.

The first page also carries a summary of the whole list: how many companies, how many you upvoted or bookmarked, the top countries and size bands — and, on a segmented list, how the companies split across its market segments. When the analyst researched the market, the first page also carries their market brief: a standalone overview of the market your list lives in, kept separate from the conversational reply so your assistant can quote or store it as its own document.

create_list and refine_list both start a real conclusia analyst run. That takes minutes, not seconds — the same run you would watch in the web app.

So they return immediately with a handle, and your assistant then checks back with get_list until the run reports itself finished, and reads the analyst's reply. While it waits, every check tells your assistant how long the run has been going, what the analyst is doing right now, how long runs like this usually take, and when to look again — so it does not have to guess. You will see the same conversation in the web app: an MCP run and a web run share one case and one history.

Two rules worth knowing:

• One run per list at a time. A second request on a list that is already running is refused rather than queued — by the time the second one started, the first would have changed the list out from under it.
• Three runs at once across all your lists. Ask for a fourth and your assistant is told to wait for one to finish. It is a safety net against a runaway loop, not a quota on how much you can do in a day.

If a run stops without finishing, the list says so and you can simply ask again. Nothing is lost.

• Anyone else's lists. Every call is checked against your own account, on every request. A list identifier from someone else's conversation does nothing.
• Archived lists — the same as in the web app.
• Anything outside the list surface: no billing, no members, no account settings, no other customers.

conclusia never receives your conversation with your assistant. It receives only the instruction you asked your assistant to send, and the one report report_problem sends when you ask for something the product cannot do.

Anything else: the exact wording of what your assistant reported helps us a lot, since every call is logged on our side with a timestamp.