Arknights Intelligence MCP

Read-only Arknights intelligence exposed over the Model Context Protocol (MCP) from one shared application core across two transports — local stdio and a private, OAuth/OIDC-protected Streamable HTTP endpoint — backed by versioned SQLite snapshots.

It answers structured stage/enemy and operator/module questions with region tagging, source provenance, and deterministic, evidence-backed analysis. User-facing tools query SQLite only; they never fetch upstream data at query time.

⚠️ Unofficial fan project. This is an unofficial, non-commercial, fan-made tool. It is not affiliated with, endorsed by, or sponsored by Hypergryph, Yostar, the maintainers of any upstream data source, Anthropic, or OpenAI. "Arknights" and all related names, logos, characters, artwork, audio, story text, and other game content are trademarks and/or copyrights of their respective rights holders. See NOTICE.

Status

v0.1 (private alpha, in development). Private and non-commercial. Public, multi-tenant hosting is explicitly out of scope and gated behind a separate readiness review — it cannot be enabled by a single configuration flag.

What it does (v0.1)

• Stage & enemy intelligence — stage metadata, enemies, routes, waves, tiles, spawns, and a deterministic stage-threat analyzer with evidence.
• Operator & module intelligence — operators, skills, talents, modules, and conservative, capability-based observations.
• Regions: en and cn. Region-specific data is never silently mixed.

What it does not do

• No query-time downloading or scraping — data comes from imported/synced snapshots only.
• No game login, no credential handling, no player-roster storage.
• No squad optimizer, combat simulator, or banner/gacha planning.
• No bundled raw snapshots or prebuilt databases in releases.

Data, licensing & policies

The Apache-2.0 LICENSE covers project code only. Imported game data and game content are separately governed — see NOTICE.

Please read, in particular:

• DATA_SOURCES.md — the source registry: owner, URL, purpose, fields consumed, region, license/permission posture, attribution, enabled state, and last review for every source.
• DATA_POLICY.md — field allowlist, excluded content, no-bulk-export policy, and provenance requirements.
• TAKEDOWN_POLICY.md — how to request attribution, correction, source exclusion, or removal, and the disable/purge/rebuild procedure.
• PRIVACY.md — remote processing, minimal logging, retention, and the no-credentials posture.
• SECURITY.md — how to report vulnerabilities and the runtime security posture.

Quickstart (local stdio)

The CLI subcommands land across milestones M1–M2; this section is the intended shape.

uv sync
# Build a local database from an approved local snapshot:
uv run arknights-mcp import --server en --source-path ./snapshot/en
# Serve over stdio for a local MCP host (Claude Code, Codex, ...):
uv run arknights-mcp serve --transport stdio

For client setup, see docs/clients/claude-code.md (Claude Code) and docs/clients/codex.md (OpenAI Codex). Both wire this server over local stdio using the current official MCP config formats. For the private remote (Streamable HTTP) transport, see docs/clients/remote.md — validating the authenticated endpoint against the MCP Inspector, Claude connector, OpenAI API, and ChatGPT web. See docs/ for architecture, adding a source, and adding a rule.

Development

uv sync
uv run ruff check .
uv run ruff format --check .
uv run mypy
uv run pytest -q

Contributor and coding-agent guardrails live in AGENTS.md and CLAUDE.md. The design of record is SPEC.md (distilled from the PRD); the founder-approved decisions in the PRD are binding and any change requires an ADR (see docs/adr/).

License

Apache-2.0 for project code only. See NOTICE for the scope limitation and trademark/attribution notices.