sugra-api-mcp

Give any AI agent access to 1,600+ data endpoints across markets, economics, companies, government, news, climate, maritime and entity screening - through one MCP server.

Works with ChatGPT, Claude, Gemini, xAI, Cursor, VS Code and any MCP client.

Official Model Context Protocol server for the Sugra API: one connector, a bundled endpoint catalog, and structured tool results with source attribution on every answer.

See it in action

An agent answering a real question end to end - resolving entities, pulling live snapshots and history, and citing the source and freshness on every number:

More examples:

Macro research - one prompt builds a full G7 inflation and policy-rate table, each cell dated and sourced, with the unavailable ones flagged rather than faked:

Cross-domain snapshot - Brent crude, marine weather and regional risk pulled together for a shipping desk, each with its source and timestamp:

What a session looks like

Hosted MCP transcript (the three composed tools shown here run on the hosted endpoint). Captured example - wording and figures vary by run and as new BLS data is published:

User: Where does US inflation stand, and how has it trended over the past year?

resolve_entity("US inflation")
  -> macro indicator cpi_us (U.S. Bureau of Labor Statistics)
get_snapshot("cpi_us")
  -> latest reading with freshness, provenance and quota cost
get_timeseries("cpi_us", metric="macro_series", range="1y")
  -> 12 monthly points with an explicit downsampling flag

Agent: US CPI printed 2.9% year over year in the latest release, down from
3.5% twelve months earlier - a steady decline since spring.
Source: U.S. Bureau of Labor Statistics via the Sugra API.

Every tool result carries structured metadata - source attribution, freshness, and rate-limit cost - so agents can cite sources and budget requests instead of guessing.

How it works

flowchart LR
    A["AI agent<br/>(ChatGPT, Claude, Gemini, xAI, IDEs)"] --> B["Sugra MCP<br/>gateway tools, plus agent tools when hosted"]
    B --> C["Sugra API<br/>1,600+ endpoints, 36 data domains"]
    C --> D["160+ primary sources<br/>markets, economics, government,<br/>news, climate, maritime"]

Behind the gateway sits the Sugra API: 160+ primary sources - sovereign statistics agencies, central banks, intergovernmental bodies and more - feeding 1,600+ endpoints across 36 data domains. The server ships a bundled catalog of the full endpoint surface, so discovery (search, describe, toolsets) runs locally without network calls; only actual data requests hit the API.

What agents build with it

The Sugra API skills live in Sugra-Systems/sugra-api-skills. The server serves five of them as MCP resources (sugra://skills/...) from a pinned commit of that repository: resources/read the URI after connect.

Agent skills

These skills teach the catalog loop. They do not add MCP tools. Connect the Sugra MCP server separately (hosted or local). The plugin package for each agent lives in Sugra-Systems/sugra-api-plugins.

Claude Code

/plugin marketplace add Sugra-Systems/sugra-api-plugins
/plugin install sugra-api@sugra-api-plugins

Skills appear as /sugra-api:<skill>, for example /sugra-api:discover-and-call.

Codex

codex plugin marketplace add Sugra-Systems/sugra-api-plugins
codex plugin add sugra-api@sugra-api-plugins

Grok

grok plugin install Sugra-Systems/sugra-api-plugins#xai

Cursor, Gemini CLI and other agents

npx skills add https://mcp.sugra.ai

Or copy the skill folders of sugra-api-skills into the agent's skills directory.

ChatGPT

The skills install from OpenAI's Plugins Directory. The MCP server attaches as a hosted connector at https://mcp.sugra.ai/mcp (permanent alias https://app.sugra.ai/mcp).

Six workflow prompts ship with the server and turn these into one-click flows in clients that surface MCP prompts:

• Market and macro research - "Compare inflation and central bank policy rates across the G7." (macro_briefing)
• Equity snapshots with sources - "Where does NVIDIA stand today - price, profile, and market backdrop?" (market_snapshot)
• Sanctions and compliance screening - "Screen this supplier and resolve its LEI identity." (sanctions_screening)
• Sector comparison - "Energy versus technology: valuations and flows side by side." (sector_compare)
• Climate, maritime and trade intelligence - "Red Sea shipping this week: chokepoint transits, crude price, and weather on the route." (earth_conditions plus the transport and commodities catalog)
• Source discovery - "What does the catalog offer for fixed income, and from which institutions?" (source_overview)

Every answer carries source attribution and freshness metadata, so agents cite instead of guessing.

Hosted MCP (recommended)

No install. In Claude, ChatGPT and Codex, add the Sugra API MCP server from a directory:

• Claude (web, desktop, mobile, Claude Code and Cowork): Add to Claude opens the Sugra API MCP server in Anthropic's Connectors Directory; connect it and sign in with your Sugra account. In claude.ai the directory is under Customize > Connectors. Claude Code signed in with a claude.ai account picks the connector up automatically; /mcp lists it.
• ChatGPT and Codex: Add to ChatGPT opens the Sugra API MCP server in the OpenAI Plugins Directory.

Any other MCP client, or a manual setup, points at the hosted Streamable HTTP endpoint:

https://mcp.sugra.ai/mcp

• The gateway tools plus the composed agent tools resolve_entity, get_snapshot and get_timeseries
• OAuth sign-in through the Claude and ChatGPT connector flows, or Authorization: Bearer sugra_xxx_... with an API key
• As a custom connector in claude.ai: Customize -> Connectors -> Add custom connector
• In ChatGPT: Settings -> Connectors -> Add MCP server

Already added Sugra to Claude as a custom connector? That connection keeps working and shows under "Custom". Connecting the Sugra API MCP server from the directory as well gives you two connections, so remove the custom one first, then connect from the directory.

Local package

Runs on your machine over stdio (or self-hosted HTTP) with an API key:

pip install sugra-api-mcp

• Eight gateway tools
• stdio for desktop clients and IDEs, Streamable HTTP for self-hosting
• Authenticates with SUGRA_API_KEY

Get a free API key at app.sugra.ai/register (Free tier: 50 req/day).

Quick start

pip install sugra-api-mcp
export SUGRA_API_KEY=sugra_xxx_...   # free key: app.sugra.ai/register
sugra-api-mcp call quotes_symbol_price --params '{"symbol":"AAPL"}'

The same call through an agent: connect the server to your client (next section) and ask "What is AAPL trading at? Use Sugra." The agent finds quotes_symbol_price in the catalog and calls it with the symbol.

Connect your client

Supported clients:

• Anthropic Claude: Claude Desktop, Claude Code (CLI), claude.ai (web)
• OpenAI GPT: ChatGPT (via MCP connector)
• Google Gemini: Gemini CLI, Gemini Code Assist (VS Code + JetBrains)
• xAI: Remote MCP Tools in xAI SDK and Responses API
• IDEs: VS Code (native), Cursor, Zed, Cline, Continue.dev, Windsurf
• Custom agents: anything built on the Python or TypeScript MCP SDK

Claude Desktop (stdio)

Add to claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: Claude Desktop has no Linux build. On Linux, pip install sugra-api-mcp and use Claude Code (CLI), an IDE client, or the hosted HTTP endpoint below.

{
  "mcpServers": {
    "sugra": {
      "command": "sugra-api-mcp",
      "env": {
        "SUGRA_API_KEY": "sugra_xxx_yourkey..."
      }
    }
  }
}

Restart Claude Desktop. Sugra tools appear in the tools menu.

Claude Code (Anthropic CLI)

Signed in to Claude Code with a claude.ai account? Add to Claude connects the Sugra API MCP server from the Connectors Directory in claude.ai, and it appears in /mcp without any local install. To run the local package instead:

claude mcp add sugra -- sugra-api-mcp
# then set the env var that sugra-api-mcp reads
export SUGRA_API_KEY=sugra_xxx_...

Or edit ~/.claude/config.json manually with the same shape as Claude Desktop above.

To install the skills as a plugin (separate from the MCP server):

/plugin marketplace add Sugra-Systems/sugra-api-plugins
/plugin install sugra-api@sugra-api-plugins

Usage with Gemini CLI

Gemini CLI reads MCP servers from ~/.gemini/settings.json (user scope) or .gemini/settings.json in the project. For a local stdio install, add:

{
  "mcpServers": {
    "sugra": {
      "command": "sugra-api-mcp",
      "env": {
        "SUGRA_API_KEY": "sugra_xxx_yourkey..."
      }
    }
  }
}

If the console script is not on PATH, use "command": "python" with "args": ["-m", "sugra_api_mcp"] instead. The equivalent Gemini CLI command is:

gemini mcp add --scope user -e SUGRA_API_KEY=sugra_xxx_yourkey... sugra sugra-api-mcp

Or connect to the hosted endpoint without installing the package:

gemini mcp add --scope user --transport http \
  --header "Authorization: Bearer sugra_xxx_yourkey..." \
  sugra https://mcp.sugra.ai/mcp

Run gemini mcp list to check the connection, then enter /mcp in an interactive session to inspect the available tools. A local stdio connection shows the eight gateway tools in Tool reference; the hosted endpoint also shows the three hosted-only agent tools. If a local server does not connect from a new directory, review and trust that workspace with gemini trust before retrying.

These examples were checked against the Gemini CLI MCP documentation and Gemini CLI v0.51.0.

Cursor, Zed, Cline, Continue.dev, Windsurf

Each of these has an MCP settings file (typically mcp.json or equivalent) with the same stdio config shape as Claude Desktop.

ChatGPT

Add to ChatGPT installs the Sugra API MCP server from the OpenAI Plugins Directory. Or add the hosted HTTP endpoint (below) as an MCP connector, since ChatGPT does not launch local stdio processes.

HTTP (claude.ai, ChatGPT, remote agents)

In claude.ai, Add to Claude connects the Sugra API MCP server from Anthropic's Connectors Directory; in ChatGPT, Add to ChatGPT installs it from the OpenAI Plugins Directory. For a manual setup or any other Streamable HTTP MCP client, use the hosted endpoint:

https://mcp.sugra.ai/mcp

Authenticate with OAuth in the connector flow or with Authorization: Bearer sugra_xxx_....

As a custom connector in claude.ai: Customize -> Connectors -> Add custom connector. In ChatGPT: Settings -> Connectors -> Add MCP server.

Tool reference

The local package exposes eight gateway tools. The hosted endpoint adds three composed analysis tools on top (see Hosted MCP above). The package exposes exactly eight tools:

Tool	Purpose
fetch_data	One-step: find best endpoint for a natural-language query and call it. Combines search + call in one round trip.
search_endpoints	Search the bundled endpoint catalog. Runtime search does not fetch /openapi.json.
describe_endpoint	Inspect an endpoint by operation_id, including path, method, parameters, required inputs, agent_hints, and request_body_schema for JSON-body POST operations.
call_endpoint	Call a Sugra API operation by operation_id. Arbitrary path calls are no longer supported.
list_toolsets	List catalog groups with endpoint counts and descriptions.
list_sources	Show bundled catalog source metadata.
sugra_entity_screen	Screen a name against sanctions and watchlists (Sugra Entity).
sugra_entity_lookup	Composed entity lookup by identifier - anchor is lei or vat, plus the identifier value; returns registry identity + screening (Sugra Entity).

call_endpoint and fetch_data both support response shaping with limit, fields, and include_raw. Shaping works on enveloped ({"data": ...}) and envelope-less payloads alike; fields entries may use dotted paths into nested objects (geo.city), and meta.shaped reports what was actually applied (fields_applied / fields_unmatched, limit_applied, records_path, order, kept_end) rather than echoing the request. limit and fields work on the records list: the envelope data list, a bare top-level array, or the one list inside an object data when exactly one of data, entries, events, history, items, observations, points, records, results, rows, series, timeseries holds a list (for example data.items on the latest news, data.observations on a FRED series). When data has no such single list but every one of its values is an object holding exactly one list named observations, as with several named sub-series side by side, limit bounds each data.<key>.observations list on its own; fields there still names keys of data. Keys beside that list, such as total and count, stay as sent, and lists nested inside records are never truncated. A fields entry that names a key of data itself projects that object instead, and a projection that matches nothing leaves the payload whole. meta.shaped.limit_applied says whether the bound took effect, and meta.shaped.records_path names the list used (data, data.<key>, data.*.observations, or null when no records list was used). limit keeps the newest end of the records list when every record carries one date or period key (such as date, period or year) in one format and the list runs one way by it: the last N records of an oldest-first list, in their order, or the first N of a newest-first list. Otherwise it keeps the first N records. Whenever a limit bounds a records list, meta.shaped.order says asc, desc or unknown and meta.shaped.kept_end says newest or first, each as a map by sub-series name for sibling sub-series. A top-level JSON array (or scalar) is always wrapped as {"data": ...} so the MCP result stays an object; otherwise FastMCP output validation reports the successful call as an error and drops the rows.

fields takes at most 32 paths, each at most 256 characters and 16 dotted parts; past any of these the call answers projection_too_large before any request is made. With fields, one projection also visits at most 100,000 list items, runs for at most 5 seconds, and takes a response of at most 2,000,000 characters of JSON before projection. That bound sits far above the 18,000-character limit on the result, which applies after projection: a 16-day weather forecast is about 295,000 characters before fields=["daily"] cuts it to fit. None of these bounds applies without fields. Shaping runs on its own pool of two worker threads, never on the event loop: at most 8 jobs run or wait for a worker, 4 of them for one caller, and a call that finds no free slot within 2 seconds answers server_busy with scope shaping or caller_shaping.

describe_endpoint returns computed agent_hints per endpoint so agents can budget time and parallelism before calling:

• duration_class - fast (under ~2s, snapshot-backed), slow (live upstream proxying, occasionally 15s+), or heavy (per-item upstream work, large batches can exceed the gateway timeout)
• max_concurrency - advisory ceiling for parallel calls from one session
• bulk_cost - on per-item bulk endpoints: 1 request credit per item in the request body (the API reports the total in the X-RateLimit-Cost response header)

Hosted-only agent tools (mcp.sugra.ai/mcp)

The hosted MCP endpoint at https://mcp.sugra.ai/mcp serves the same eight tools PLUS three composed agent tools that are not available on stdio or self-hosted installs:

Tool	Purpose
resolve_entity	Free text (ticker, company, indicator, coin, currency pair) to a canonical market or macro entity. Ambiguous matches return ranked candidates, never a silent pick.
get_snapshot	Entity plus a named recipe to one composed current view with freshness, provenance, coverage, and billing blocks. Composed calls charge a fixed recipe cost (1-2 requests) from the daily quota.
get_timeseries	Entity plus metric (price, macro_series, etf_flows, etf_monthly_flows) to a bounded series with an explicit downsampling flag. etf_flows estimates at filing cadence; etf_monthly_flows is the fund's own NPORT-P monthly creations and redemptions.

These three tools wrap an internal composed plane that requires an infrastructure credential available only on the hosted deployment. The tool code ships inside the package, but it is registered only by the hosted HTTP entry point and only when that credential is present - pip install sugra-api-mcp (stdio and self-hosted HTTP) always exposes the classic eight-tool gateway. Hosted-only examples in any documentation are labeled as such. For compliance entity lookups (LEI / VAT, sanctions screening) use sugra_entity_lookup and sugra_entity_screen, which work on every transport.

CLI

Server startup is unchanged:

sugra-api-mcp
sugra-api-mcp --transport streamable-http --port 8001

Catalog and gateway helpers:

sugra-api-mcp doctor
sugra-api-mcp list-toolsets
sugra-api-mcp search "NASDAQ futures"
sugra-api-mcp describe cot_financial
sugra-api-mcp call quotes_symbol_price --params '{"symbol":"AAPL"}'

Environment variables

User-facing configuration for local installs, MCP clients, Docker stdio, and directory sandboxes (for example Glama Try in Browser). Set only this:

Variable	Required	Default	Description
SUGRA_API_KEY	For API calls	-	Your Sugra API key (sugra_...). Get a free key at app.sugra.ai/register (Free tier: 50 req/day). Not needed to start the server: catalog tools (search_endpoints, describe_endpoint, list_toolsets, list_sources) work without it; API-calling tools return a structured missing_api_key error until it is set. In HTTP mode with a client Bearer token this is only a fallback.

Optional overrides (leave unset unless you need them):

Variable	Default	Description
SUGRA_API_BASE	https://sugra.ai	Override the Sugra API base URL (self-hosted or beta API only).
SUGRA_TIMEOUT	30	Downstream HTTP timeout in seconds for calls from this server to the Sugra API.

Operator-only settings for self-hosted Streamable HTTP (reverse proxy CORS/hosts, OAuth authorization-server wiring, and shared secrets) are documented in docs/self-hosting.md. Do not put operator secrets into public directory sandboxes. SUGRA_MCP_TRUST_PROXY_HEADERS (off by default) makes the server read X-Real-IP and X-Forwarded-Host; set it only behind a reverse proxy that overwrites both on every request, and never where clients can reach the process directly (details in the same guide). SUGRA_MCP_LIMITS (off by default) turns on request limits counted in the server's own memory; the same guide lists its settings.

HTTP transport with OAuth

When running with --transport streamable-http the server allows unauthenticated MCP discovery requests (initialize, notifications/initialized, tools/list, resources/list, prompts/list, and ping) so ChatGPT Apps and other mixed-auth clients can discover tool metadata. Tool calls still require Authorization: Bearer .... Two token formats are accepted:

• Raw API key (sugra_...) - passed through as the downstream x-api-key. Compatible with earlier local API-key setups.
• OAuth JWT - signature verified against the issuer's JWKS. The audience must match https://app.sugra.ai/mcp, the token must include sugra:read, and hosted access is validated against APP before resolving the user's primary API key. Successful hosted OAuth requests update MCP connection activity in APP.

Most users should use the hosted endpoint https://mcp.sugra.ai/mcp with an API key as Bearer instead of self-hosting OAuth; standalone OAuth clients sign in at the alias https://app.sugra.ai/mcp for now. If you run your own HTTP process, see docs/self-hosting.md.

Timeouts and the error contract

SUGRA_TIMEOUT caps each downstream HTTP call from this server to the Sugra API (default 30 seconds). It is one link in a longer chain; when a tool call fails, elapsed_ms in the error payload tells you which link cut it:

MCP client (agent harness)         own tool timeout, often 60-180s, client-controlled
  -> hosted proxy (app.sugra.ai)   86400s, effectively unlimited
    -> this server (httpx)         SUGRA_TIMEOUT, default 30s
      -> Sugra API -> upstreams    15-60s per upstream call, server-side

Tool failures return structured JSON instead of raising, so agents can pick a retry strategy: