MCP App Store
by KeenEthics
Lucius icon

Lucius MCP Server for ChatGPT

By Lucius AI Systems, Inc.

Overview

Lucius is tender intelligence for companies that bid on public contracts. This plugin gives ChatGPT the Lucius tender index and its tender reading. Search open public tenders from official sources, including EU TED, UK Find a Tender and Contracts Finder, Public Contracts Scotland, SAM.gov, Grants.gov, the World Bank, the Inter-American Development Bank, South Africa's eTenders, CanadaBuys and Ireland's eTenders. Filter by words, country, source, CPV code or how soon a tender closes. Each result carries the buyer, country, deadline, value where published, a short description and the link to the buyer's own notice, with the source's attribution. Open one tender's details, or find tenders similar to one you have. Ask Lucius to analyse a tender from your search, or send a public link to a tender pack, and Lucius reads it: every requirement it finds, tied to the page it was read on, with the forms to fill and a bid or no-bid call. Each analysis uses one of the workspace's monthly analyses, like an upload in the Lucius app. You need a Lucius account; a workspace owner or admin approves the connection. Lucius never submits a bid, edits a tender or sends anything on your behalf.

How to connect

  1. ChatGPT: Open the Lucius listing in the ChatGPT app directory and choose Connect. ChatGPT listing Lucius exposes 6 tools here.

Tools

analyze_pack

ChatGPT
Analyse a tender pack. Give either tender_id, the id of a tender from search_tenders, get_tender or similar_tenders (Lucius fetches that notice's own documents; when its own links hold none, the documents of the same tender's notice on another source, matched by the buyer's file reference, or the buyer's portal the notice names, and documents_from says which), or url, a public link to a tender document (PDF, Word, Excel or ZIP) or to a notice page that links to the documents. Lucius fetches it, starts the analysis in the background and returns an analysis_id at once. Poll get_pack_analysis with that id until its status is complete or failed. The analysis is the one the Lucius web app runs on an upload: it uses one of the workspace's monthly analyses under its Lucius plan, it is stored in the workspace (the team sees it in the app too), and only this workspace's API keys can read it. Links on portals whose terms do not allow Lucius to pass their material on (UNGM, GeBIZ, Uganda eGP, Zambia e-GP, GETS, CPPP, AusTender) are refused; download the files and send them to POST /api/v1/ext/packs over REST instead, which is also how to send files you hold. A tender this workspace has already analysed, or the same link or file sent again within 24 hours, returns the existing analysis (deduplicated: true) and uses no analysis. A new verdict or judge_as setting is a new analysis. By default the bid call is judged against this workspace's company profile; to read a tender for a prospect, send judge_as (a short description of the prospect) or verdict "none" (no call).

ask_market

ChatGPT
Answer an aggregate market question in plain words over the Lucius AI index of public procurement notices and award records: counts, rankings, totals, averages, trends, and who won at what price, across many notices. Lucius turns the question into SQL, runs it on its public index and returns a short answer, the SQL that ran (sql) and the result table (columns and rows, at most 25 rows; rows_cut says when there were more). Show the SQL and the table with the answer. Not for finding notices to bid on or reading one notice: use search_tenders and get_tender for those. Notices cover 2026 only. Award data covers World Bank contract awards since 2015, EU contract awards of every sector published on TED since May 2023, Spanish PLACSP vaccine and medicine results, and UK Contracts Finder awards. Each answer carries an attribution, the credit the sources' licences ask for; keep it with the figures. When the question is ambiguous the result is a clarifying_question to put to the user instead of a table. Takes up to 45 seconds.

get_pack_analysis

ChatGPT
Status and result of a pack analysis started with analyze_pack (or through the REST API, or in the Lucius app by this workspace). status is queued, processing, complete or failed; a failed analysis carries the reason in plain words. When complete, result holds what the analysis produced: title, buyer, deadline, the bid or no-bid call and why, the requirements with their original text quoted from the pack and the document and page each was read on, which requirements decide the bid (deciding: true, the pass/fail gates first), and the forms the pack asks bidders to return when the pack's forms have been mapped. A finished analysis is not always a bid call: result.document says what was read (classification, kind, is_tender), and when is_tender is false (a bid, a signed contract, something else, or a document Lucius could not classify) verdict is null, no requirement is deciding and document.note says why. When the workspace has not described its company there is no call either: verdict.call is null, verdict.open_requirements counts the deciding requirements still unanswered and verdict.note says what to do. result.coverage says whether every file got the full read (complete), every limit of this read in plain words (notes), how each file of a pack was read (files: full, text, partial or not_read) and whether a second read is still adding requirements (second_read: pending, with requirements_final false until it ends). Requirements come in pages (requirements_limit, at most 1000; requirements_next_offset for the next page); deciding_only returns just the deciding ones. On a plan without the full requirements list only the deciding requirements are returned and requirements_locked is true. The first page (requirements_offset 0) also carries price_schedule, the buyer's price form or bill of quantities as Lucius listed it (lines, quantities, units and pages, no prices), and spec_register, the requirements sorted item by item into a technical specification register. Each has a status: ready, none (the pack has none), stale (built from an earlier requirements list), pending, failed, null (not built yet) or locked (the workspace's plan does not include them for this tender). An analysis started with verdict "none" has verdict null; one started with judge_as has its verdict judged for that company, with judged_as and judged_as_label ("judged as: ..."). Either way result.verdict_mode says how it was judged (absent for the default, a call judged against this workspace). Polls that answer queued, processing or failed are free; reading a complete result is one request.

get_tender

ChatGPT
One public procurement notice by id: its metadata, snippet and source link, plus your company's stored go/no-go analysis (landing summary and bid verdict) when your company has analysed that notice in Lucius. Also documents: the documents and links the source's own feed lists for the notice (bidding documents, attachments, the notice page, a buyer's portal), each as the address on the publisher's side, with the type, format, date and name where the source gives them (at most 20). Lucius serves no file. Today Find a Tender, Contracts Finder and SAM.gov list them; SAM.gov gives the links alone. The list is empty for the other sources. Data: public procurement notices from the Lucius AI catalog (TED, Find a Tender, Contracts Finder, Public Contracts Scotland, SAM.gov, Grants.gov, World Bank, Inter-American Development Bank, South Africa National Treasury eTenders, Plataforma de Contratación del Sector Público, CanadaBuys, eTenders, European Commission, Funding & Tenders Portal). Each result is notice metadata (title, buyer, country, deadline, value with its currency where the source states one and value_label saying what the amount is, such as an estimated total value or a budget excluding VAT, CPV codes, source), a snippet of at most 700 characters and a link to the source notice; the full notice text is not served. Each result also carries an attribution, the credit its source's licence asks for; some sources' terms allow no snippet, so their snippet is null. Licences: https://ailucius.com/developers#licences

search_tenders

ChatGPT
Search open public procurement notices by free text (semantic and keyword, fused) with optional country, source, CPV and closing-window filters. With no query it lists open notices by closing date. Returns a page of notices and a cursor for the next page. A tender published on two sources (a national portal and TED) is listed once, as the copy whose source lists its documents; also_on names the other copy, with its own value. Data: public procurement notices from the Lucius AI catalog (TED, Find a Tender, Contracts Finder, Public Contracts Scotland, SAM.gov, Grants.gov, World Bank, Inter-American Development Bank, South Africa National Treasury eTenders, Plataforma de Contratación del Sector Público, CanadaBuys, eTenders, European Commission, Funding & Tenders Portal). Each result is notice metadata (title, buyer, country, deadline, value with its currency where the source states one and value_label saying what the amount is, such as an estimated total value or a budget excluding VAT, CPV codes, source), a snippet of at most 700 characters and a link to the source notice; the full notice text is not served. Each result also carries an attribution, the credit its source's licence asks for; some sources' terms allow no snippet, so their snippet is null. Licences: https://ailucius.com/developers#licences

similar_tenders

ChatGPT
Open public procurement notices most similar to a given notice, by stored embedding (no model call), optionally limited to countries. Data: public procurement notices from the Lucius AI catalog (TED, Find a Tender, Contracts Finder, Public Contracts Scotland, SAM.gov, Grants.gov, World Bank, Inter-American Development Bank, South Africa National Treasury eTenders, Plataforma de Contratación del Sector Público, CanadaBuys, eTenders, European Commission, Funding & Tenders Portal). Each result is notice metadata (title, buyer, country, deadline, value with its currency where the source states one and value_label saying what the amount is, such as an estimated total value or a budget excluding VAT, CPV codes, source), a snippet of at most 700 characters and a link to the source notice; the full notice text is not served. Each result also carries an attribution, the credit its source's licence asks for; some sources' terms allow no snippet, so their snippet is null. Licences: https://ailucius.com/developers#licences

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/lucius"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/lucius)