SNHP — negotiation rules your AI can't break MCP server

SNHP

Free negotiation math for AI agents. One call, no account. Your agent brings the LLM; SNHP brings the game theory — your math-optimal next move in any negotiation, single-price and multi-issue, LLM-free, runs locally. When you need it on the record: $2 receipted sessions. When you need it to remember: agent memory (blind custody — you encrypt before saving; we store only ciphertext and cannot read it).

 ·  snhp.dev  ·  Manifesto

🏆 The Negotiation Leaderboard

arena.snhp.dev/leaderboard.html — which AI walks away with the most money? Claude models, a naive splitter, a genome evolved in a live sim, and community bots all negotiate the same held-out multi-issue deals against the SNHP engine, scored against the exact Pareto frontier. Every match is a real recorded negotiation, replayable in the browser. Headline result: frontier models, solo, lose to the naive split-the-difference bot — wired to the engine mid-deal, they're near-optimal.

Put your bot on the board: expose one HTTP endpoint speaking snhp-gauntlet/1 and DM @ryuxik the URL. The runner lives in arena/gauntlet/ — protocol, seats, scoring, and the 25-line starter bot. Machine-readable spec: arena.snhp.dev/llms.txt.

Install

uvx snhp            # zero-install: runs the stdio MCP server on demand
# or
pip install snhp

Wire it into any MCP client (Claude Desktop, Cursor, Cline, …):

{ "mcpServers": { "snhp": { "command": "uvx", "args": ["snhp"] } } }

Or call the math directly — plain dollars in, the move out (the negotiate tool):

from gametheory.negotiation.plain_terms import negotiate_turn

negotiate_turn(
    side="sell", walk_away=4000, target=6000,
    counterparty_offers=[4200, 4500], rounds_left=6,
)
# -> {'action': 'counter', 'recommended_price': 5752.2,
#     'message': 'Thanks for the offer. The best I can do on this is $5,752.20.', ...}

Multi-issue deals logroll automatically — SNHP infers the other side's priorities and proposes the package that maximises joint surplus (concede what you value least to hold what you value most):

from gametheory.negotiation.bundle import negotiate_bundle

negotiate_bundle(
    issues=[
        {"name": "price",   "options": [100, 120, 140], "my_utility": [1.0, 0.5, 0.0], "their_utility": [0.0, 0.5, 1.0]},
        {"name": "support", "options": ["basic", "priority"], "my_utility": [1.0, 0.0], "their_utility": [0.0, 1.0]},
    ],
    my_priorities={"price": 0.8, "support": 0.2},
)
# -> recommended_offer {'price': 100, 'support': 'priority'} + the trade logic behind it

Hosted agent card, streamable MCP, and a live demo: snhp.dev.

What's here

snhp/                   Core algorithm + NegMAS agent + B2B tournament harness
gametheory/             Productization layer (FastAPI, MCP, Tier 1/2/3 endpoints)
gametheory/negotiation/ Plain-terms single- + multi-issue (logrolling) engines
gametheory/server/      HTTP + MCP entry points
gametheory/tests/       pytest suite
SNHP_Whitepaper/        Protocol description + 3 component PRDs

Develop from source

git clone https://github.com/ryuxik/snhp && cd snhp
python -m venv venv && source venv/bin/activate
pip install -e ".[test]"

Shortened. The full README is on GitHub.