gambit-mcp
# gambit-mcp
Pre-production MCP server for bounded, read-only Gambit quantitative-research
workflows. The target protocol revision is `2026-07-28`; MCP Python SDK `2.2.0`
also serves the tested `2025-11-25` legacy era.
```sh
make setup
make check
uv run gambit-mcp
```
M1 is complete and M2 market-data contract groundwork is underway:
- `gambit://server/about` and `gambit://capabilities` resources;
- `gambit_check_contract`, a synthetic schema/compatibility tool;
- no market data, Gambit calculations, persistence, remote authorization, live
trading, arbitrary code, filesystem paths, or arbitrary network access.
No market-data tool is exposed yet. The immutable Gambit artifact, ingestion
format, parser-isolation evidence, and data/quant semantic approvals remain M2
gates; see `docs/M2_READINESS.md`.
Set `APP_TRANSPORT=streamable-http` to run the pre-production HTTP transport on
localhost. It is not approved for remote production use.
Configuration uses `APP_` environment variables documented in `.env.example`.
Production refuses to start with an untraceable revision. The inherited HTTP
health and container deployment contract must be adapted and requalified for the
MCP ASGI app before container promotion. See `PROJECT_BRIEF.md`,
`docs/PROJECT_PLAN.md`, and `docs/MCP_SERVER_ENGINEERING_STANDARD.md`.
TDQS
Scored across 1 tool
With only one tool, there is no possibility of confusing it with another. The description clearly states exactly what it validates and what it does not, leaving no ambiguity.
The name 'gambit_check_contract' follows a clear verb_noun pattern and is readable. However, with only one tool, there is no broader naming convention to assess, and the redundant 'gambit_' prefix is slightly awkward.
A single tool feels thin for a server, even one with a narrow scope. It is usable for its stated purpose but provides agents with a very limited surface.
The server's stated purpose is limited to checking contract compatibility for a pre-production adapter, and the tool fully covers that purpose. Missing capabilities like data validation or backtesting are explicitly out of scope, so they are not true gaps, but the surface is still minimal.