pdata Prediction Markets MCP server

pdata-client

Python client, AI-assistant plugin files and sample data for pdata.world, a free, read-only API for prediction-market data from eight venues: Polymarket, Kalshi, Manifold, Myriad, Limitless, Predict, Opinion and Gemini.

pip install pdata-client

from pdata import Client

pd = Client()
page = pd.markets(source=["kalshi", "polymarket"], closed=False, sort="-volume_24hr", page_size=5)
for m in page["items"]:
    print(m["source"], f"{m['probability']:.0%}", m["event_title"], "—", m["question"])

Cite as: Data aggregated by pdata.world (https://pdata.world) from 8 prediction-market venues; CC BY 4.0; source data © each venue. Dataset DOI 10.5281/zenodo.23211610.

pdata.world polls each venue every 10 to 60 minutes and keeps prices, 24-hour volume and results in one schema, so one request can compare venues that otherwise each need their own integration. There is no API key and no signup. The data is published under CC BY 4.0: use it, and credit pdata.world.

Install

pip install pdata-client
# or the latest commit:
pip install git+https://github.com/noskillcoding/pdata-client

The client uses only the standard library (urllib, json, gzip), has no dependencies and runs on Python 3.9 or later. You can also copy src/pdata/client.py and src/pdata/_version.py into a project. It identifies itself as pdata-client/<version> (+https://pdata.world/agents); pass user_agent= to name your own project instead.

The MCP server and the assistant plugins

pdata.world hosts an MCP server at https://api.pdata.world/mcp (Streamable HTTP, no key, no account). It is read-only: its 14 tools read the same public data as the API. The server is run by pdata.world and its code is closed-source; it is not in this repository. This repository holds the Python client, the plugin files that connect coding assistants to the hosted server, and notes on the datasets. The server is listed in the official MCP registry as world.pdata/mcp.

• Claude Code: /plugin install pdata --marketplace noskillcoding/pdata-client inside a session (Claude Code 2.1.275 or later), or from the shell claude plugin marketplace add noskillcoding/pdata-client followed by claude plugin install pdata@pdata-world. The plugin adds the MCP server and a skill, skills/pdata-prediction-markets/SKILL.md, that tells the assistant when to use pdata and how to cite it.
• Gemini CLI: gemini extensions install https://github.com/noskillcoding/pdata-client adds the same server and skill.
• Any other MCP client: add the server by URL, for example claude mcp add --transport http pdata https://api.pdata.world/mcp, or the config in examples/mcp.json.

What is in this repository

Path	What it is
src/pdata/	The Python client, published on PyPI as pdata-client
.claude-plugin/, .mcp.json, skills/	Claude Code plugin and its one-plugin marketplace
gemini-extension.json	Gemini CLI extension (uses the same skills/)
apify/	Source of an Apify actor that returns events with current prices
data/	Samples of the nightly bulk files, CC BY 4.0
examples/	Runnable scripts and an MCP client config

What the client covers

Method	API operation	What it returns
markets(**filters), iter_markets(...)	GET /markets	Markets, filtered and sorted, one page or all pages
market(source, id)	GET /markets/{source}/{id}	One market
top_movers(...), closing_soon(...)	GET /markets/top-movers, /closing-soon	Largest 24h price changes; markets closing next
events(**filters), iter_events(...)	GET /events	Events (an event groups the markets of one question)
event(source, id)	GET /events/{source}/{id}	One event with its markets, or its final results once archived
event_history(source, id, range=...)	GET /events/{source}/{id}/history	Price and 24h-volume series, 24h / 7d / 30d / all
similar_events(source, id)	GET /events/{source}/{id}/similar	The same or similar question on other venues
search(q)	GET /search/suggest	Events and markets by title
summary()	GET /summary	Per-venue counts and 24h volume
news_movers(...)	GET /news/movers	The price moves behind pdata.world/news
bulk_index(), iter_bulk(kind)	/bulk/	Nightly full files, streamed without saving

Every method returns the API's JSON unchanged, so the API reference describes every field. Unknown filters are rejected with a message listing every allowed one. Errors raise PdataError with the API's code and message.

Runnable examples are in examples/: quickstart.py makes five common calls, and resolved_archive.py streams the archive of settled events.

Other ways to get the same data

• MCP server for AI assistants: https://api.pdata.world/mcp; see above.
• Nightly bulk files at https://api.pdata.world/bulk/: every open market (markets-open), every open event (events-open) and the archive of settled events with their results (resolved), as gzipped JSON lines. Line 1 of each file is a manifest giving the licence and every field's meaning and unit.
• Monthly snapshot on Hugging Face: the same three files as Parquet, each snapshot tagged with its date: lowskillcoding/pdata-prediction-markets.
• Read-only Postgres with full SQL: postgresql://pdata_readonly:pdata_public_2026@pg.pdata.world:6433/pdata_new. The password is published on purpose and is never rotated. Queries are capped at 5 seconds. See examples/postgres.sql.
• The website: every event has a page at https://pdata.world/events/{source}/{id}, and https://pdata.world/data describes all of the above.

Sample data

data/ holds random samples of the three bulk files as of 7 October 2026: 1,000 open markets, 500 open events and 1,000 settled events, with an equal share per venue where the venue had enough rows. Each file starts with the manifest line. They are for looking at the shape of the data; the full files are rebuilt every night.

Units and things to know

• probability and prices are on a 0–1 scale.
• volume and volume_24hr are in each venue's own unit and are not converted across venues: shares for Polymarket and contracts for Kalshi (each pays $1 at resolution), play-money mana for Manifold, and each other venue's figure as it reports it. How eight venues count volume explains each one.
• Price history starts in May 2026. Points older than 7 days are kept as 12-hour buckets, and a market's history is deleted 10 days after it closes. After that, the event's URL returns "kind": "tombstone" with each top market's final result.
• In the resolved archive, each market's probability is the last price pdata saw, which is usually after the market settled. It records the result; it is not a forecast.
• Events that closed between 2 June and 11 August 2026 are missing from the resolved archive: they were pruned before a backup was taken.
• pdata groups Kalshi's sibling events for one game under a single group event (is_group); its volume is the sum of its children.

Research using this data

pdata.world publishes dated reports built from this data, each with its method and limits:

• How accurate were prediction-market prices a day before the result? For 39,681 Polymarket and Kalshi markets that settled between 28 September and 5 October 2026, grouped into 10-point bands by their price 24 hours before close, the share that resolved YES was on average 0.96 percentage points from the band's average price (Brier score 0.110, against 0.199 for always guessing the base rate). Also published as a citable note: doi.org/10.5281/zenodo.23212130.
• How eight prediction-market venues count volume
• What each prediction-market API gives you
• Same question, different price: prediction markets across venues
• What eight prediction-market venues list, by category

Citing

The data is CC BY 4.0. A credit such as "Data: pdata.world" with a link is enough. Every single-event API answer also carries _meta.cite_as, a ready-made citation sentence. For papers, cite the dataset on Zenodo: DOI 10.5281/zenodo.23211610 always resolves to the latest snapshot, and each snapshot has its own DOI, such as 10.5281/zenodo.23211611 for 7 October 2026. The calibration note is 10.5281/zenodo.23212130. GitHub's "Cite this repository" button uses CITATION.cff.

Licence

The code in this repository, including the plugin files and the Apify actor, is under the MIT licence. The data, including the files in data/, is under CC BY 4.0; the underlying market data remains subject to each venue's own terms.

pdata.world is independent and not affiliated with any of the venues it tracks.

Releasing

One version covers the PyPI package, the Claude Code plugin and the Gemini extension. Bump it in src/pdata/_version.py, .claude-plugin/plugin.json and gemini-extension.json together (a test fails otherwise), then push a tag v<version>: publish.yml builds, tests and uploads to PyPI through trusted publishing. Contact: support@pdata.world.