actually-mcp-server
Enables AI agents to check news against live Polymarket markets, retrieve market odds, and manage trades including placing, selling, and canceling orders.
README
Actually - What Markets Really Think
A Chrome extension and an MCP server that put prediction-market odds on top of the news you are already reading.
News tells you that "experts are concerned" or that "odds are rising". It rarely tells you a number. The number exists - on a prediction market, where people are betting their own money on the outcome. It just lives in another tab, and finding it means knowing such a market exists in the first place.
Actually closes that gap. Open an article, click the toolbar icon, and the extension matches the text against live Polymarket markets and shows the market's own probability. No account, no wallet, no signup.

What it does
-
Matches an article to a market. A local embedding model (
Xenova/all-MiniLM-L12-v2, ~33 MB) scores the article text against the cached market set, boosted by keyword and number overlap - numbers matter because the encoder is nearly blind to them, and "dip to $57,500" embeds almost identically to "reach $120,000". -
Runs on your device. Model and WASM runtime are bundled at build time - nothing is fetched from a CDN at install or runtime, and on the default settings the article text never leaves the machine.
-
Reads non-English pages. They are translated to English before matching, using Chrome's own built-in
TranslatorAPI (local, no network). Every Polymarket question is written in English, and the local model holds 86 Cyrillic tokens out of 30522 - an untranslated Russian headline embeds to noise. Translated, the same headline scores 0.53-0.71 against the market it is actually about, against a 0.35 floor.
-
Trades, optionally. Connect any WalletConnect v2 wallet for limit and market orders, selling, cancelling resting orders, and a positions panel with cost basis and P&L. Keys and funds are never held by us.
-
Serves agents too. The same capabilities are exposed over the Model Context Protocol:
check_news,get_market,place_order,sell_order,cancel_order,get_positions.
How it fits together
extension/ Chrome MV3 extension (React popup, offscreen document)
extension/worker/ Cloudflare Worker - API proxy, rate limiting, market cache
packages/core/ Matching, pricing and Polymarket API logic, shared by both clients
packages/mcp-server/ MCP server, published to npm as actually-mcp-server
packages/market-cache-builder/ Cron job that precomputes the market embeddings
The heavy work lives in an offscreen document because MV3 service workers cannot run WASM, hold a WebSocket, or survive long enough to sign an order.
The market cache is 2000 open markets with precomputed embeddings, rebuilt every two hours. Selection blends three orderings - 24h volume, lifetime volume, and recency - because ranking by lifetime volume alone is the wrong shelf for a news tool: a market opened this morning under today's headline has no history and loses to a year-old election market every time.
The Worker exists so no client has to hold a credential. It proxies Polymarket's APIs, serves the precomputed cache, and acts as a remote signer for the relayer, so the builder credential stays on the server and never ships inside an extension.
Running it
npm install
npm run models:fetch -w extension # bundle the embedding model
npm run build -w extension # type-check + build to extension/dist
Load extension/dist as an unpacked extension in Chrome. Copy
extension/.env.example to .env.local and fill it in first - the build bakes
the Worker URL, the WalletConnect project id and the builder code.
npm test --workspaces # 686 tests across four workspaces
Privacy
Discovery is free and needs no account. There are no content scripts: the page is read only when you click, and only the active tab. On the default settings (local embeddings, translation off-device) the article text does not leave your machine. Telemetry is opt-in and off by default.
Full policy: https://actually-api.sofiaseremeteva.workers.dev/privacy
Status
The extension is built and passing its release gates; the Chrome Web Store
submission is in progress. actually-mcp-server is published on npm.
Redeeming a resolved position is in testing: builder authentication, neg-risk contract selection and the zero-balance guard are each verified against live services, but no redeem has yet been observed collecting funds end to end.
Trading is unavailable in several jurisdictions (US, GB, FR, BE, AU, SG, TH, TW, PL, and sanctioned countries), enforced Worker-side. Viewing odds works everywhere.
Actually does not provide financial advice.
License
See LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.