Provides AI agents with tools for DFS settlement, betting odds math, expected value calculations, and sports entertainment predictions.
Also known as Buzzr-app/dfs-engine, dfs-engine, @buzzr/mcp
- Stars
- 1
- Tools
- 11
- Downloads, 30 days
- 220
- Latest version
- 5.1.0
- Last commit
- Sep 13, 2026
- License
- MIT
What is Buzzr Sports Engine MCP?
Buzzr Sports Engine provides AI agents with tools for DFS settlement, betting odds math, expected value calculations, and sports entertainment predictions. It exposes 11 tools, such as grade_dfs_entry, grade_dfs_entries, validate_dfs_entry.
Summary generated from the project's README
About Buzzr Sports Engine MCP server
Buzzr Sports Engine provides a specialized toolset for AI agents to perform complex sports-related calculations and settlements. It functions as a local stdio MCP server that connects to clients like Claude Desktop, Claude Code, Cursor, and Codex. The server acts as a thin, schema-validated interface for three core engines: DFS settlement, betting math, and entertainment predictions. It does not fetch live odds or private user data, instead relying on the inputs provided by the agent.
Developers can use the 11 available tools to handle tasks such as calculating Kelly stakes, removing vig from two-way markets, or settling daily fantasy sports entries. The DFS engine supports specific policies for platforms like PrizePicks and Underdog, allowing for the validation and grading of pick-em entries. The entertainment engine uses machine learning to provide game buzz scores and personalized rankings. This server is designed for local execution via npx and requires Node.js 22 or higher.
Written from the project's README. Every use case is checked against a quote from it.
How to install Buzzr Sports Engine MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall Buzzr Sports Engine MCP in Claude Code
$ claude mcp add dfs-engine -- npx -y @buzzr/mcpInstall Buzzr Sports Engine MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"dfs-engine": {
"command": "npx",
"args": [
"-y",
"@buzzr/mcp"
]
}
}
}Install Buzzr Sports Engine MCP in VS Code
.vscode/mcp.json
{
"servers": {
"dfs-engine": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@buzzr/mcp"
]
}
}
}Install Buzzr Sports Engine MCP in VS Code CLI
$ code --add-mcp '{"name":"dfs-engine","type":"stdio","command":"npx","args":["-y","@buzzr/mcp"]}'Install Buzzr Sports Engine MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"dfs-engine": {
"command": "npx",
"args": [
"-y",
"@buzzr/mcp"
]
}
}
}Install Buzzr Sports Engine MCP in Codex
$ codex mcp add dfs-engine -- npx -y @buzzr/mcpInstall Buzzr Sports Engine MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.dfs-engine] command = "npx" args = ["-y", "@buzzr/mcp"]
Install Buzzr Sports Engine MCP in Gemini CLI
$ gemini mcp add -s user dfs-engine npx -- -y @buzzr/mcpInstall Buzzr Sports Engine MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"dfs-engine": {
"command": "npx",
"args": [
"-y",
"@buzzr/mcp"
]
}
}
}Buzzr Sports Engine MCP tools 11
Detected from the READMEgrade_dfs_entrygrade_dfs_entriesvalidate_dfs_entrylist_book_policiesfair_lineclosing_line_valueparlay_valuekelly_stakesummarize_bet_historypredict_game_buzzrank_games
| grade_dfs_entry | — | |
| grade_dfs_entries | — | |
| validate_dfs_entry | — | |
| list_book_policies | — | |
| fair_line | — | |
| closing_line_value | — | |
| parlay_value | — | |
| kelly_stake | — | |
| summarize_bet_history | — | |
| predict_game_buzz | — | |
| rank_games | — |
Automated checks
Ran Nov 14, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachablebuzzr-app/dfs-engine is public and reachablegithub.com/buzzr-app/dfs-engine ↗Pass
- Package names this repository as its source@buzzr/mcp names buzzr-app/dfs-engine as sourcenpmjs.com/package/@buzzr/mcp ↗Pass
- Build provenance (Sigstore attestation)Sigstore attestation for @buzzr/mcpnpmjs.com/package/@buzzr/mcp/v/5.1.0#provenance ↗Pass
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-09-13github.com/buzzr-app/dfs-engine ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from @buzzr/mcpPass
- No critical advisory in the package or its dependencies102 resolved dependencies checked; no critical advisoryosv.dev ↗Pass
- No hidden instructions to the model11 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHub@buzzr/mcp
MCP server exposing the @buzzr sports engines to AI agents — DFS settlement, odds math, and entertainment predictions as tools.
@buzzr/mcp is a Model Context Protocol stdio
server that puts the whole @buzzr engine family in front of any MCP-capable agent
(Claude Desktop, Claude Code, or your own client). All math and policy logic lives
in the underlying engines — this package is a thin, schema-validated tool surface:
@buzzr/dfs-engine— pick-em settlement with versioned compatibility and custom policies@buzzr/bets-core— no-vig fair lines, parlay pricing, expected value, Kelly staking@buzzr/entertainment-engine— ML buzz-score predictions and personalized game ranking
Install
Run it directly with npx (Node 22+). Pin the version you reviewed instead of
silently accepting a future latest release:
npx -y @buzzr/mcp@5.1.0
The server speaks MCP over stdio: JSON-RPC on stdin/stdout, logs on stderr. It exposes 11 tools with bounded inputs and outputs, and does not fetch live odds, box scores, operator accounts, or private user data.
Smithery
The Buzzr Sports Engine listing on
Smithery packages
the exact published @buzzr/mcp@5.1.0 artifact as a local stdio MCPB. It is a
distribution and installation surface, not a hosted HTTP service. To add it to
Codex through Smithery:
npx -y smithery@1.2.0 mcp add sarveshsea/buzzr-sports-engine --client codex
Smithery writes the corresponding MCP configuration for the selected client. Use the direct npm command above when you need both the server package and its launcher configuration pinned explicitly.
Claude Desktop
Add this to claude_desktop_config.json from Settings → Developer → Edit
Config, then fully quit and reopen Claude Desktop:
{
"mcpServers": {
"buzzr": {
"command": "npx",
"args": ["-y", "@buzzr/mcp@5.1.0"]
}
}
}
Claude Code
claude mcp add --transport stdio buzzr -- npx -y @buzzr/mcp@5.1.0
For a version-controlled project configuration, add this to .mcp.json, trust
the project when prompted, and start a new Claude Code session:
{
"mcpServers": {
"buzzr": {
"command": "npx",
"args": ["-y", "@buzzr/mcp@5.1.0"]
}
}
}
Confirm it with claude mcp get buzzr; inside Claude Code, /mcp shows the
connection and discovered tool count.
Cursor
Add this to the repository's .cursor/mcp.json (or the equivalent user-level
MCP configuration), then fully restart Cursor:
{
"mcpServers": {
"buzzr": {
"command": "npx",
"args": ["-y", "@buzzr/mcp@5.1.0"]
}
}
}
Open Cursor Settings → MCP to confirm buzzr connected and exposed 11
tools. See Cursor's MCP documentation if
your installed Cursor version presents a different settings location.
Codex
The Codex CLI, IDE extension, and app share config.toml. The one-command user
setup is:
codex mcp add buzzr -- npx -y @buzzr/mcp@5.1.0
Or add the equivalent block to ~/.codex/config.toml for all projects, or to a
trusted repository's .codex/config.toml for that project only:
[mcp_servers.buzzr]
command = "npx"
args = ["-y", "@buzzr/mcp@5.1.0"]
Run codex mcp get buzzr, then start a new Codex task after changing the
configuration.
Generic stdio MCP client
Use this process configuration in any client that accepts a command plus args:
{
"name": "buzzr",
"transport": "stdio",
"command": "npx",
"args": ["-y", "@buzzr/mcp@5.1.0"],
"env": {}
}
Treat stdout as JSON-RPC only, read diagnostics from stderr, and close the child
process's stdin during shutdown. The client must perform the MCP lifecycle below;
starting npx in a terminal and seeing it remain open is normal for a stdio
server waiting for a client.
Initialization and capability discovery
The client sends initialize first. Verify the serverInfo.name field is
"buzzr", the serverInfo.version field is the installed package version, and
the response has a capabilities.tools object. Send
notifications/initialized, then call tools/list. The server returns the 11
tools in the catalog below with their input schemas. It does not advertise data
fetching, resources, or prompts.
examples/mcp-calls.json is a machine-readable JSON-RPC
transcript with the initialization expectation, exact tool discovery order, and
a safe list_book_policies → validate_dfs_entry → grade_dfs_entry workflow.
The repository replays that workflow through a real MCP client in CI.
Startup troubleshooting
- Confirm Node.js 22+ and npm are visible to the same desktop process or shell:
node --version,npm --version, andnpx --version. - Confirm the pinned release exists with
npm view @buzzr/mcp@5.1.0 version, then runnpm cache verify. If npm reports cache corruption, repair npm's cache before retrying;npm cache clean --forceis a last resort because it removes the whole local cache. - GUI apps may inherit a smaller
PATHthan an interactive shell. On macOS or Linux, runcommand -v npx; on Windows, runwhere.exe npx. If the client cannot findnpx, use that absolute path ascommand(normallynpx.cmdon Windows) or launch the client from an environment where Node is onPATH. - Keep logs off stdout. A healthy manual start writes a
listening on stdiodiagnostic to stderr and waits for JSON-RPC input. - After any config, Node, PATH, or package-version change, fully quit and restart Claude Desktop, Claude Code, Cursor, or Codex. Opening only another chat tab may leave the old MCP child process and cached tool catalog in place.
Tool catalog
| Tool | Engine | What it does |
|---|---|---|
grade_dfs_entry |
@buzzr/dfs-engine | Settle one transport-bounded 1–12-leg entry; executable built-in policies currently allow at most 6 PrizePicks or 8 Underdog legs. |
grade_dfs_entries |
@buzzr/dfs-engine | Settle 1–50 entries, up to 600 total legs, with bounded concurrency and isolated failures. |
validate_dfs_entry |
@buzzr/dfs-engine | Return structured engine validation issues for a candidate entry without settling it. |
list_book_policies |
@buzzr/dfs-engine | List authoritative executable profile snapshots and metadata-only drafts, including status, verification, sources, and complete play types. |
fair_line |
@buzzr/bets-core | Remove vig from both sides of one two-way market. |
closing_line_value |
@buzzr/bets-core | Compare placed and closing prices for the same selection. |
parlay_value |
@buzzr/bets-core | Price independent parlay legs, compare offered odds, and optionally calculate expected value. |
kelly_stake |
@buzzr/bets-core | Calculate full and fractional Kelly stakes from a supplied win probability. |
summarize_bet_history |
@buzzr/bets-core | Summarize up to 500 bets with overall and UTC-period rollups, drawdown, and streaks. |
predict_game_buzz |
@buzzr/entertainment-engine | Predict one game's 1–10 entertainment score with confidence and factor detail. |
rank_games |
@buzzr/entertainment-engine | Rank 1–100 candidate games for a bounded taste profile. |
Tool strings, identifiers, arrays, stdio frames, concurrent calls, and serialized
results are bounded. American odds must be within [-100000, -100] or
[100, 100000]. grade_dfs_entries, closing_line_value, and
summarize_bet_history return string contractVersion: "1".
DFS policy safety
Operator-named policies are independent compatibility profiles, not official rules engines or evidence of affiliation:
- PrizePicks is experimental and partially verified. Standard payout references were reviewed on 2026-07-16 from PrizePicks Payouts and PrizePicks Potential Outcomes, with DNP behavior checked against DNPs, Reboots, and Ties. The compatibility policy treats a 2-pick Power entry with a DNP as a refund when it falls below the two-pick minimum. Other settlement behavior and variable lineup-specific payouts remain incomplete.
- Underdog is experimental and unverified. The Underdog Sports Legal Center is the recorded rules entrypoint; the current compatibility payout and settlement values have not been verified.
Displayed lineup terms are authoritative. Call list_book_policies before
grading, inspect verification and sources, and obtain explicit operator rulings
for DNPs, reboots, ties, rescues, voids, and corrections. The tool lists future
fixtures with executable: false; grading tools reject draft book IDs because
drafts are metadata, not settlement implementations.
Error contracts
- Invalid tool arguments return the same bounded
invalid_inputresult through an MCP client transport or a direct exportedtool.handler(...)call. At most eight compact validation issues are included; raw Zod errors are never returned. - Malformed JSON-RPC envelopes remain protocol errors owned by the MCP SDK and
are distinct from a valid
tools/callrequest with invalid tool arguments. validate_dfs_entryintentionally accepts a bounded candidate object and returns the engine's structured validation report.- Execution failures return generic
isError: trueresults such astool_execution_failed,entry_settlement_failed,server_busy, orresult_too_large. Internal error details are not public.
Example transcripts
"Did my PrizePicks slip hit?"
User: I had a $10 2-pick power play: LeBron over 25.5 points (he scored 31) and Curry over 27.5 (he scored 33). What's the payout?
Agent calls
grade_dfs_entry:{ "entryId": "slip-1", "bookId": "prizepicks", "playTypeId": "power", "stake": 10, "displayedMultiplier": 3, "legs": [ { "legId": "l1", "playerName": "LeBron James", "league": "NBA", "propType": "points", "line": 25.5, "direction": "over", "actual": 31 }, { "legId": "l2", "playerName": "Stephen Curry", "league": "NBA", "propType": "points", "line": 27.5, "direction": "over", "actual": 33 } ] }Result:
"status": "won","payout": { "total": 30, "withdrawable": 30, "bonus": 0 }— both supplied actuals clear their lines and the submitted 3× displayed multiplier is consistent with the selected compatibility table. Confirm the actual entry details before treating this as an operator outcome.
"Is this parlay +EV?"
User: The book offers +264 on a two-leg parlay where both legs are -110/-110 markets. Worth $100?
Agent calls
parlay_valuewith{ "legs": [{ "selected": -110, "opposite": -110 }, { "selected": -110, "opposite": -110 }], "offeredAmericanOdds": 264, "stake": 100 }.Result: fair probability 0.25 (fair price +300) vs. offered +264 —
"edgePercent": -2.47and an expected loss of about $9 per $100. The agent recommends passing.
"What should I watch tonight?"
User: I'm a Lakers fan. Rank tonight's slate for me.
Agent calls
rank_gameswith the slate and{ "profile": { "favoriteTeams": ["Lakers"], "leagueAffinity": { "NBA": 0.8 } } }.Result: each game comes back with
baseScore, boundedaffinityAdjustmentandsocialAdjustment, atotalScore, and named factors ("favorite team playing", "league affinity") the agent can cite when explaining the ranking.
Embedding
The server is also exported as a library, so you can mount the same tool catalog on your own transport (in-memory for tests, HTTP, etc.):
import { createBuzzrMcpServer, allTools } from '@buzzr/mcp';
const server = createBuzzrMcpServer();
await server.connect(myTransport);
Individual tool definitions (gradeDfsEntryTool, fairLineTool, …) are exported
too — each is { name, title, description, inputSchema, handler }, and handlers
can be called directly without any transport.
Codex skill
The repository includes a Buzzr Sports Engine skill with the 11-tool routing guide, limits, response-reading order, and operator-safety rules:
npx skills@1.5.17 add https://github.com/Buzzr-app/dfs-engine --skill buzzr-sports-engine --agent codex --yes --copy
Repository contributors can prove the same one-command install without using a
published branch: npx skills@1.5.17 add . --skill buzzr-sports-engine --agent codex --yes --copy. npm run check:skill runs the official pinned
quick_validate.py, performs that local install in an isolated temporary
project, and verifies every installed skill file byte-for-byte.
Compatibility
- Node.js >= 22 is supported.
rank_gamesrequires@buzzr/entertainment-engine>= 5.0.0. Against an older engine build the tool degrades gracefully with anengine_capability_missingerror result instead of crashing the server.
Import the supported API from @buzzr/mcp. Deep src/* and dist/* imports are unsupported.
See the all-package API index and the
generated root-export reference.
Report reproducible defects in GitHub Issues. Report vulnerabilities privately through SECURITY.md. The versioning and support policy defines the supported runtime and SemVer contract.
License
Buzzr Sports Engine MCP questions
How do I add Buzzr Sports Engine MCP to Claude Code?
Run claude mcp add dfs-engine -- npx -y @buzzr/mcp in your terminal. Add --scope user to make it available in every project.
How do I add Buzzr Sports Engine MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What tools does Buzzr Sports Engine MCP provide?
It exposes 11 tools, including grade_dfs_entry, grade_dfs_entries, validate_dfs_entry, list_book_policies, fair_line.
What kind of betting math tools are included?
The server includes tools for removing vig from two-way markets, pricing parlays, calculating expected value, and determining Kelly stakes.
Does this server fetch live sports data?
No, the server does not fetch live odds, box scores, operator accounts, or private user data.
Can I use this with Claude Desktop?
Yes, you can add the server to your Claude Desktop configuration by editing your claude_desktop_config.json file.
What are the limitations on DFS entry settlement?
Built-in policies currently allow settling at most 6 PrizePicks legs or 8 Underdog legs per entry.