polymarket-research
Read-only MCP server for investigating Polymarket wallets, markets, and cross-wallet patterns using 21 tools backed by public APIs and an optional SQLite cache.
README
polymarket-research
Read-only MCP server that gives Claude Code investigative access to Polymarket — wallet forensics, market metadata, and cross-wallet pattern analysis. No order placement, no signing, no private keys.
All four tiers ship: 21 tools backed by Polymarket's public Gamma + Data + CLOB APIs, a local SQLite cache, and (optionally) Polygon RPC for on-chain enrichment.
Architecture
┌──────────────┐ stdio ┌──────────────────┐ HTTPS ┌─────────────────────┐
│ Claude Code │ ─────────► │ MCP server │ ──────────► │ gamma / clob / data │
│ (or any MCP │ │ (FastMCP, 21 │ │ polymarket APIs │
│ client) │ ◄───────── │ tools) │ ◄────────── │ (public, no auth) │
└──────────────┘ └──────┬───────────┘ └─────────────────────┘
│
▼ (cache + cohort queries)
┌─────────────┐ ┌────────────────────┐
│ SQLite │ opt-in (env) │ Polygon RPC │
│ (~/.poly…) │ ───────────────► │ (Tier-4 chain │
└─────────────┘ │ enrichment only) │
└────────────────────┘
Pure-read: the server never signs or moves funds. Trades are stored
append-only in SQLite by (transaction_hash, log_index); profiles are TTL'd.
Polygon RPC is fully opt-in via POLYGON_RPC_URL — without it, Tiers 1–3 run
end-to-end on the public HTTP APIs alone.
Tier 1 tools — wallet investigation
| Tool | What it does |
|---|---|
get_wallet_profile |
Identity + total portfolio value (cached, TTL 1h). |
get_wallet_positions |
Open positions, sorted by current USD value. |
get_wallet_trades |
Trade history, newest first; paginates Data API. |
get_wallet_pnl_breakdown |
Cash PnL grouped by category / month / market. |
get_wallet_top_trades |
Largest trades by notional USD; wins/losses filter. |
get_wallet_concentration |
HHI, Gini, top-5 share, biggest single market. |
get_wallet_first_trade |
Earliest trade — proxy for "join date". |
Tier 2 tools — markets & events
| Tool | What it does |
|---|---|
get_market |
Single market by id / conditionId / slug. Caches payload. |
get_market_orderbook |
CLOB book snapshot for one outcome (mid, spread, depth). |
get_market_holders |
Top holders across a market's outcome tokens. |
get_market_price_history |
CLOB price history for one outcome (1m / 1h / 1d / …). |
search_markets |
Filter Gamma markets by activity, slug substring, sort. |
get_event |
Event + child markets by slug or id (e.g. "GTA VI" group). |
Tier 3 tools — cross-wallet
| Tool | What it does |
|---|---|
correlate_wallets |
Jaccard + Pearson on bucketed trade timelines; common markets. |
find_whale_followers |
Lag percentiles for candidates that copy a leader on same market+side. |
find_market_cohort |
All wallets in the cache that traded a given market (with rollups). |
get_cache_stats |
Counters of cache contents — trades, traders, markets, etc. |
Bulk loader (scripts/ingest_historical.py)
Cohort tools query the local SQLite cache, so populate it first either by running Tier-1 tools against the wallets you care about, or by bulk-loading a historical dataset (CSV / JSONL / JSON):
venv/bin/python scripts/ingest_historical.py \
--file ~/data/polymarket_trades.csv --format csv \
--col-tx-hash transactionHash --col-wallet proxyWallet \
--col-condition-id conditionId --col-ts timestamp
--dry-run, --since 2024-01-01, --limit N, and --db-path are all
supported. Re-running is idempotent (PK is (tx_hash, log_index)).
Tier 4 tools — forensics
| Tool | What it does |
|---|---|
find_data_source_concentrators |
Cluster active markets by resolver/source; rank by aggregate volume. |
compare_entry_to_consensus |
Compare a trade's entry vs CLOB consensus mid at trade time (bps deviation). |
get_wallet_alpha_decay |
Bucket a wallet's trades on a market into deciles of its lifetime; surfaces early-vs-late edge. |
get_wallet_timing_pattern |
UTC hour-of-day histogram + best-effort timezone inference. |
compare_entry_to_consensus accepts an optional enrich_with_chain=True flag
that looks up the trade's block timestamp via Polygon RPC. That path is opt-in:
set POLYGON_RPC_URL and install the chain extra (pip install -e '.[chain]').
Without it, the tool runs purely off CLOB price history.
Setup
git clone <this repo>
cd polymarket-mcp
uv venv venv --python python3.12
uv pip install --python venv/bin/python -e .
(python3 -m venv venv works too if you have python3-venv installed.)
Environment
| Var | Default | Purpose |
|---|---|---|
POLYMARKET_GAMMA_BASE |
https://gamma-api.polymarket.com |
Override Gamma host. |
POLYMARKET_CLOB_BASE |
https://clob.polymarket.com |
Override CLOB host (Tier 2). |
POLYMARKET_DATA_BASE |
https://data-api.polymarket.com |
Override Data host. |
POLYGON_RPC_URL |
(unset) | Polygon RPC for Tier 4 forensics. |
POLYMARKET_RATE_LIMIT_RPS |
5 |
Per-host token-bucket cap. |
POLYMARKET_DB_PATH |
~/.polymarket_research.db |
SQLite cache file. |
POLYMARKET_CACHE_MODE |
warm |
warm / bypass / offline. |
Run the server standalone
PYTHONPATH=src venv/bin/python -m polymarket_research.server
You should see the FastMCP banner; the process is now listening on stdio.
Connect via Claude Code
A ready-to-go .mcp.json ships at the repo root. From inside Claude Code,
either start it from this directory or symlink the file. Then restart the
MCP layer (or run /mcp restart polymarket-research) and the seven tools
above appear under polymarket-research:*.
Recipes
Investigate a wallet:
> Use polymarket-research:get_wallet_profile on 0x<wallet>.
> Then get_wallet_concentration and pnl_breakdown by category.
Spot a high-conviction position:
> get_wallet_top_trades 0x<wallet> by notional_usd, n=10, side=losses
> compare those trades to get_wallet_pnl_breakdown by market
Find a wallet's earliest activity:
> get_wallet_first_trade 0x<wallet>
Replace <wallet> with any Polymarket trader address — use search_markets
get_market_holdersto discover candidates on markets that interest you.
Cache
Trades are stored append-only by (transaction_hash, log_index); wallet and
market profiles are TTL'd. Inspect manually:
sqlite3 ~/.polymarket_research.db ".schema"
sqlite3 ~/.polymarket_research.db "SELECT COUNT(*) FROM trades"
Reset the cache by deleting the file. The schema is recreated on next run.
Tests
PYTHONPATH=src venv/bin/python -m pytest tests/ -v --cov=polymarket_research
126 tests, ~90% coverage on parsers/analytics. Heavy coverage on the pure
layer; endpoints exercise fakes for HTTP and :memory: SQLite for cache.
Read-only invariant
This server never imports a private key, never signs, never places an order. Verify with:
grep -rEn 'sign|private_key|eip712|order_place|submit_order|wallet_create' src/
(Any hits should be substrings inside docstrings or comments — never live code.)
Disclaimer
This server reads only public Polymarket and Polygon on-chain data. It does not sign, place orders, or move funds. Outputs (concentration, alpha decay, timing patterns, cohort overlaps) are research signals — not financial advice. Use of the tool implies acceptance that all analyzed data is public and that the author is not affiliated with Polymarket.
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.