MCP server
The Model Context Protocol, the open standard for handing tools to AI agents
Your cited data,1 in any AI agent.
Point Claude Code, Claude Desktop or any MCP client at one endpoint, and your agent pulls quotes, scores, screens and reference data, each answer stamped with the moment it was measured.
POSThttps://convexityos.com/api/v1/mcpStreamable HTTP
1Every data result is dated by its newest point, never by the time of the request: meta.data_as_of on the data tools, asOf on a Copilot run.
- tools/callget_convexity_score, answered with its as-of time
- tools/list59 read-only tools, filed by scope
Connect
The same bearer key as the REST API
Two lines, and the round trip begins.
01Claude Codeclaude mcp add
claude mcp add --transport http convexity https://convexityos.com/api/v1/mcp \
--header "Authorization: Bearer YOUR_KEY"02Any MCP client.mcp.json
{
"mcpServers": {
"convexity": {
"type": "http",
"url": "https://convexityos.com/api/v1/mcp",
"headers": { "Authorization": "Bearer YOUR_KEY" }
}
}
}The sessionsample
- →
initializefree - →
tools/listfree - >
get_convexity_score("NVDA") - ←
200 DataEnvelope data.ridge_score_pct87data.sector_etfXLKdata.window_days180meta.data_as_of2026-07-21T03:01Zmeta.sync_statushealthy
“As of 2026-07-21, NVDA sits in the 87th percentile.”The agent cites, instead of guessing.
Authentication is the same bearer key as the REST API, sent as an Authorization header. Mint one from /settings/developer; the key shows its plaintext value once, so store it before closing the panel.
Tools
59 read-only, 4 scopes
59 read-only tools, scoped per key.
data:read50portfolio:read6quote:read2score:read1
- 01
get_quotequote:readCurrent quotes for up to 25 tickers. - 02
get_price_historyquote:readOHLCV history for one ticker, ranges 1D through ALL. - 03
get_convexity_scorescore:readLatest nightly Convexity Score: the forecast, its percentile among covered names, and its rank in its sector. - 04
get_company_referencedata:readCompany profile, key stats, valuation fields. - 05
run_screenerdata:readScreen US equities by score, sector, market cap, valuation ratios, insider buys, and more. - 06
get_company_eventsdata:readUpcoming earnings plus recent actuals vs estimates. - 07
get_related_companiesdata:readPeers, sector members, and supply-chain counterparties.
Each tool requires the scope beside it, and a key without it gets an error the agent can read, naming the scope to mint. No write tools, no order placement, no trade execution: the server reads, it never writes. The seven above are a representative slice; all 59, with request and response shapes, are in the API reference.
Honest
Metering, freshness, surface
It meters like the REST API. It dates its data. It cannot write.
Metering
Every tool call draws against the same per-minute and per-day rate limits and the same monthly cap as the REST API. The handshake (initialize, tools/list) is free: only real data requests are metered.
Freshness
The data tools return a DataEnvelope whose meta.data_as_of is computed from the data itself, not request time, so the agent cites “as of 2026-07-21” instead of presenting stale data as current. A Copilot run answers in a shape of its own, carrying its own asOf and sources, and the ontology manifest, a registry defined in code, carries a version instead.
Surface
Every tool reads; there is no write path and no trade execution. Most need only public-data scopes; the portfolio tools read the key owner's own book and nothing else, and still cannot write or trade. The worst a leaked key can do is read data within its scopes until you revoke it from the dashboard.
Give your agent market data it can cite.
API access is included with Developer ($99/mo) and Quant, the two tiers that carry the programmatic stack. Mint a key, paste two lines, and Convexity is a tool your agent can call.