Ihalent
FreeNot checkedTurns Turkish public-tender result notices into structured records and provides analytics on firm history, discount distributions, and competition metrics via a
About
Turns Turkish public-tender result notices into structured records and provides analytics on firm history, discount distributions, and competition metrics via an MCP server.
README
Turkish public-tender award intelligence: who won, for how much, at what discount, against how many bidders.
Reads the result notices KİK publishes in the daily Kamu İhale Bülteni, and anything you collected earlier with saidsurucu/ihale-mcp; ihalent structures and analyzes them.
Every public tender in Turkey ends with a Sonuç İlanı — a result notice that states the estimate, the contract price, the winner, and how many firms bid. The notices are public. They are also one at a time, unstructured, and impossible to reason across: you cannot ask EKAP "what has this company won in the last two years" or "how far below estimate does this authority actually award." ihalent turns those notices into structured records and answers exactly those questions — firm histories, discount (kırım) distributions, and competition metrics — with every number traceable back to the notice it came from.
That is real output over nine real construction awards from the December-2025 / January-2026 window (in
examples/). The -2.96% at the bottom of the range is not a bug: Istanbul
University awarded a 615-million-lira campus job at 2.96% above its own estimate,
through an emergency "pazarlık" procedure — a contract signed over the public estimate,
which is precisely the kind of thing that should be easy to see and currently is not.
Why this exists
I'm an industrial engineer and I work in construction. When our firm weighs a public tender, the questions that decide the bid are all about other people's history: how far below estimate does this authority award, how many firms usually show up, what has a given competitor been winning and at what price. All of that is public — it is sitting in tens of thousands of result notices on EKAP — and none of it is queryable. So everyone rebuilds a private, partial version of it by hand, in spreadsheets, badly.
The notices themselves are published in full every working day, in the Kamu İhale Bülteni. ihalent is the layer that turns that pile of notices into an answer: it reads the bulletin (or notices you collected earlier with saidsurucu/ihale-mcp), structures them, and does the analytics that don't exist yet.
What it answers
A firm's history — every award across the dataset, folded across spelling variants:
$ ihalent firm examples/sample-awards.jsonl "ATAŞ MÜHENDİSLİK"
ATAŞ MÜHENDİSLİK HARİTA İMAR PLANLAMA İNŞAAT GIDA HAYVANCILIK SANAYİ VE TİCARET
LİMİTED ŞİRKETİ
wins: 1 joint ventures: 1 total contract value: 19.709.997,40 TL
Discount (kırım) — ATAŞ MÜHENDİSLİK
HARİTA İMAR PLANLAMA İNŞAAT GIDA
HAYVANCILIK SANAYİ VE TİCARET LİMİTED
ŞİRKETİ
┌────────┬────────┬────────┬────────┐
│ mean │ median │ min │ max │
├────────┼────────┼────────┼────────┤
│ 33.87% │ 33.87% │ 33.87% │ 33.87% │
└────────┴────────┴────────┴────────┘
based on all 1 awards.
top awarding authorities:
1x DSİ 14. Bölge Müdürlüğü
Where the discounts are — grouped and sorted, highest first:
$ ihalent discounts examples/sample-awards.jsonl --by authority
Mean discount by authority
┌──────────────────────────────────┬────────┬────────┬────────┐
│ authority │ mean │ median │ awards │
├──────────────────────────────────┼────────┼────────┼────────┤
│ DSİ 14. Bölge Müdürlüğü │ 33.87% │ 33.87% │ 1/1 │
│ Ziraat Bankası Genel Müdürlüğü │ 31.41% │ 31.41% │ 1/1 │
│ İstanbul YİKOB │ 26.98% │ 26.98% │ 1/1 │
│ İSKİ Genel Müdürlüğü │ 26.66% │ 26.66% │ 1/1 │
│ İTÜ Rektörlüğü │ 22.68% │ 22.68% │ 1/1 │
│ Esenler Belediyesi │ 21.99% │ 21.99% │ 1/1 │
│ Ağrı İl Özel İdaresi │ 19.99% │ 19.99% │ 1/1 │
│ Kağıthane Belediyesi │ 17.41% │ 17.41% │ 1/1 │
│ İstanbul Üniversitesi-Cerrahpaşa │ -2.96% │ -2.96% │ 1/1 │
└──────────────────────────────────┴────────┴────────┴────────┘
awards column: used / considered (excluded lack an estimate, or were awarded in lots).
Competition — how many valid bids show up, and how often exactly one does (a single-bid award is a flag procurement watchdogs care about). It's in the overview above.
The one rule
A number is never shown without the ground it stands on. A mean discount always comes
with "over how many awards, and how many were dropped for a missing estimate." A firm's
total contract value says so when some of its wins had no published price, so you read it as
a floor, not the full figure. A discount is None, never 0, when the estimate wasn't
published — ihalent does not invent the numbers the government didn't print. Half the value
of a tool like this is refusing to guess.
Install
pip install ihalent # the core
pip install "ihalent[bulletin]" # + reading the Kamu İhale Bülteni PDFs
pip install "ihalent[mcp]" # + the MCP server
Or from source: pip install git+https://github.com/gulmezeren2-byte/ihalent.
The workflow
ihalent reads a JSONL file of awards. The way to produce one is the Kamu İhale Bülteni: KİK prints every result notice in a daily bulletin (Law 4734, Art. 47), one file per tender type, on EKAP's public bulletin page. ihalent downloads those files and reads them:
# 1. collect: today's works (yapım) bulletin, or every working day since a date
# (earlier days go through the page's archive form)
ihalent fetch-bulletin --out bulletins
ihalent fetch-bulletin --out bulletins --since 2026-11-01
# 2. structure: always over the whole folder (see "Lots" below)
ihalent ingest-bulletin bulletins -o awards.jsonl
# 3. ask:
ihalent overview awards.jsonl
ihalent firm awards.jsonl "ACME İNŞAAT"
ihalent discounts awards.jsonl --by authority --min 5
ingest-bulletin checks every file against the bulletin's own table of contents and
says how many notices it read, rejected, or could not find; it exits 1 if those don't
add up. A notice whose labels and values don't pair up is rejected, never guessed.
Every working day, by itself. ihalent daily FOLDER (0.5.0 and later; it needs the
bulletin extra) does both steps for every tender type: it fetches what the folder is
missing from the last seven days, so a week the computer was off is filled in, and
rebuilds FOLDER/awards.jsonl. Each file is read once and its read is kept beside it in
.ihalent-reads/, so the run stays quick as the folder grows. Let the computer's
scheduler run it after the day's bulletin is out:
# Windows (PowerShell): every weekday at 19:00, or as soon as the computer is on after that
$ihalent = (Get-Command ihalent).Source
$run = New-ScheduledTaskAction -Execute $ihalent -Argument "daily `"$HOME\ekap`""
$when = New-ScheduledTaskTrigger -Weekly -DaysOfWeek Monday,Tuesday,Wednesday,Thursday,Friday -At 19:00
$opts = New-ScheduledTaskSettingsSet -StartWhenAvailable -AllowStartIfOnBatteries -DontStopIfGoingOnBatteries
Register-ScheduledTask -TaskName ihalent-daily -Action $run -Trigger $when -Settings $opts
# macOS / Linux: add this line with `crontab -e` (`command -v ihalent` prints the full path)
0 19 * * 1-5 /full/path/to/ihalent daily "$HOME/ekap"
A public holiday just shows as a day with no bulletin; a whole week without one exits 1,
because that is also how a changed page would look. If the page ever asks for human
verification or refuses, daily stops and leaves ihalent-stopped.txt in the folder;
later runs don't contact the page until you delete that file.
Lots. A tender awarded in parts (kısmi teklif) gets one notice per lot, and the lots
come out on different days, each printing the tender-wide estimate. ihalent merges them
by İKN and sums the contracts, but gives such a tender no discount
(lots_unverified): no bulletin says when the last lot has been printed, so comparing
the estimate with the lots seen so far would invent a number. Each lot's winner and
price are kept (lots), so a firm is credited only the lots it won.
What the bulletin does not carry: a province field. --by province stays empty for
bulletin data; ihalent does not infer it. Amounts in another currency: some goods
notices price the contract in euros or dollars against a lira estimate. That contract's
lira price is left empty, with no exchange rate guessed, so it gets no discount; an
estimate printed in lira with its foreign original beside it is read in lira. What was
verified: the works and goods bulletins of 12 and 13 November 2026 and the service and
consultancy bulletins of the 13th — 883 of 883 notices read, each file matching its table
of contents. All were fetched with fetch-bulletin: through the archive form (the two
goods days in one --since run) and, for the works bulletin of the 13th, through the
day's own button. A notice that says it has no estimate ("- TRY", or a sentence saying
so) is read with the estimate left empty.
The older path, via ihale-mcp. Since September 2026 EKAP's search API asks for human
verification, so collecting with ihale-mcp
from your own machine fails with HTTP 428. ihalent does not try to get past that check.
If you still have notices collected that way, ihalent ingest bundle.json reads them.
Every command takes --json for pipelines and agents. Or skip collection and try the
bundled example directly:
git clone https://github.com/gulmezeren2-byte/ihalent && cd ihalent
python examples/build_sample.py # parses the real notices in examples/notices/
ihalent overview examples/sample-awards.jsonl
Commands
| command | what it does |
|---|---|
ihalent overview AWARDS |
value, discount, competition and the data gaps of a dataset |
ihalent firm AWARDS "NAME" |
one company's wins, total value, discount, and where it wins |
ihalent discounts AWARDS --by X |
mean/median discount grouped by authority, province or tender_type |
ihalent single-bid AWARDS |
awards with a single valid bid — no real competition (a watchdog flag) |
ihalent concentration AWARDS [--authority X] |
winner concentration (HHI) — do the same few firms win everything? |
ihalent flags AWARDS |
per-award red flags: single bid, near-estimate price, no estimate, high drop-off |
ihalent fetch-bulletin --out DIR |
download Kamu İhale Bülteni files from EKAP's public bulletin page |
ihalent ingest-bulletin DIR |
bulletin result notices → awards JSONL, checked against the table of contents |
ihalent daily DIR |
the last week's bulletins of every type → DIR, then DIR/awards.jsonl; for a scheduler |
ihalent parse NOTICE.md |
one result notice → structured JSON |
ihalent ingest BUNDLE.json |
collected ihale-mcp/EKAP output → awards JSONL |
Using it with AI agents
The result notice is unstructured text; the interesting questions are aggregate. That is an
awkward fit for an agent working notice-by-notice, and a natural fit for a tool: --json
output with stable fields, an exit code that means something, and a firm-name match that
folds spelling variants so an agent doesn't have to.
There is a native MCP server (pip install 'ihalent[mcp]') that exposes the analytics
as tools — overview, firm, discounts, concentration, flags, parse_notice,
ingest_bundle — over a dataset you point it at:
IHALENT_AWARDS=awards.jsonl ihalent-mcp
No local Python? The Dockerfile builds the same server:
docker build -t ihalent . && docker run --rm -i -e IHALENT_AWARDS=/data/awards.jsonl -v "$PWD:/data:ro" ihalent.
Point it at the awards built from your bulletins and an agent can reason across them in
one session. The Python API (ihalent.bulletin.read_bulletin,
ihalent.bulletin.awards_from_reads, ihalent.analytics.firm_profile) is three calls deep
if you'd rather script it.
Scope, honestly
- One narrow network step, and it never forces its way in. Everything except
fetch-bulletinanddailyis a pure function of the files you give it. They ask one public page for the files KİK publishes there for anyone: one file per working day and tender type, at least two seconds apart, never twice. If that page ever asks for human verification or refuses a request, they stop and say so — they do not retry (dailyleaves a file that keeps later runs from trying), and ihalent will not ship a way around such a check. - Company-name folding is conservative on purpose. It merges legal-form suffixes and
Turkish spelling variants, and it would rather show two rows for one firm than one row for
two — so
firmreports how many distinct spellings a query matched, and warns you if that is more than one. - The parser tracks one document: the result notice. Bid-level detail (who else bid, at what price) is not in the notice and so is not here. Discount is the estimate-to-contract gap, which the notice does carry.
- Award values are nominal lira, as printed. Comparing 2023 and 2026 contracts is your analysis to make with the dates in hand; ihalent does not silently inflation-adjust.
What ihalent is not
It is not a scraper of EKAP's search, and not a way around its checks. It is not a live
dashboard or a paid tender-alert service; it is a library and a CLI you point at data you
control. And it does not redistribute a dataset: it ships the
handful of real notices in examples/ and the code to structure your own.
A note on the data
Public-tender results are public information, published by the state for public scrutiny. ihalent structures what EKAP already discloses. Most winners are companies acting in their public commercial capacity — but some are sole traders, whose name is a private person's name and personal data under Turkish law (KVKK). So ihalent never keeps the winner's address and ships no collected bulletins or awards. It downloads the bulletins to your computer and reads them there; it publishes nothing. What you then keep, use or share, sole traders' names included, is your responsibility under the law, not this project's.
How this project is built
I designed the model and the analytics and I review every line; I use AI agents (Claude Code) heavily for implementation speed, and the commit trailers say so. The contract is the tests — 132 of them, built on the exact JSON shapes EKAP and ihale-mcp emit and on the page geometry measured in a real bulletin, including a result notice with a negative discount, a cancelled tender, a value that runs onto the next page and a tender awarded in lots. They don't care who typed them.
Related
- acikpoz — the other half of the Turkish public-procurement picture.
ihalentreads the tender result notices that say who won and at what discount;acikpozreads the unit-price (birim fiyat) catalogs that say what the work should have cost. Same discipline: every figure traceable to its source, nothing invented.
More tools by Eren Gülmez.
License
MIT — Mehmet Eren Gülmez
Installing Ihalent
This server has no published package — it is built from source. Open the repository and follow its README.
▸ github.com/gulmezeren2-byte/ihalentFAQ
Is Ihalent MCP free?
Yes, Ihalent MCP is free — one-click install via Unyly at no cost.
Does Ihalent need an API key?
No, Ihalent runs without API keys or environment variables.
Is Ihalent hosted or self-hosted?
Self-hosted: the server runs locally on your machine via the install command above.
How do I install Ihalent in Claude Desktop, Claude Code or Cursor?
Open Ihalent on unyly.org, pick your client tab (Claude Desktop, Claude Code, Cursor) and press Install — the config is generated automatically, no JSON editing.
Related MCPs
GitHub
PRs, issues, code search, CI status
by GitHubFilesystem
Secure file operations with configurable access controls.
Memory
Knowledge graph-based persistent memory system.
Template MCP Server
A CLI tool to create a new Model Context Protocol server project with TypeScript support, dual transport options, and an extensible structure
by mcpdotdirectServer Sequential Thinking
MCP server for sequential thinking and problem solving
by GitHub ActionsUnyly Deploy
Deploy sites and MCP servers to Unyly straight from your AI agent
by UnylyAmap Maps Mcp Server
MCP server for using the AMap Maps API
by duxiaohuiPayMCP
Lightweight payments layer for MCP servers: turn tools into paid endpoints with a two-line decorator. [PyPI](https://pypi.org/project/paymcp/) · [npm](https://w
by blustAIServer For Power BI
Enables exploring Microsoft Fabric / Power BI workspaces and semantic models, and executing ad-hoc DAX queries.
by Mitsubishi-FusoGpt Image 2 Mcp
RunAPI GPT Image 2 MCP server for image generation: create tasks, poll results, and check pricing across 1 model from Claude Code, Codex, Cursor, and VS Code.
by runapi-aiCompare Ihalent with
Not sure what to pick?
Find your stack in 60 seconds
Author?
Embed badge for your README
Browse similar
All development MCPs
