alphaswarm-ml-mcp

Catalog date: 2026-06-24.

Dedicated MCP server publishing the data.ml.* tool slice — Predictor Hub lookups, AlphaBacktestExperiment dispatch, walk-forward run inspection, finetune trainer status, model serving (vLLM / Ollama / KServe). Piggybacked on the alphaswarm-core pod (same FastAPI app, distinct route prefix and aud claim).

This is the MLOps slice's RFC 9728 / RFC 8707 conformant endpoint — see the mcp-rfc-conformance rule in the private alphaswarm_internal repo. (As of 2026-07-19, alphaswarm/.cursor/rules/ only carries a pointer to this SSoT repo.)

Identity

Field	Value
Service id	alphaswarm-ml-mcp
Role	mcp
Package	alphaswarm_models/ (tools); served from alphaswarm/ml_mcp/
Image (key)	(piggybacked on api)
Built from	alphaswarm_platform/Dockerfile (target api)

Wire

Field	Value
Protocol	HTTP/1.1 + WebSocket (MCP)
Port	8000 (shared with alphaswarm-core)
Health	GET /mcp/ml/tools (lists tool registrations)
Discovery	GET /.well-known/oauth-protected-resource/mcp/ml (RFC 9728 metadata)
Audience claim	dedicated per-MCP aud per AGENTS rule 49

Tool registrations

Tool prefix	Concept doc
data.ml.models.*	ml-framework.md
data.ml.deployments.*	ml-framework.md
data.ml.skills.*	mlops-service.md
data.ml.serving.*	ml-framework.md
data.ml.hetegnn.*	analysis-framework.md
data.ml.predict, data.ml.forecast, data.ml.classify, data.ml.segment, data.ml.analyze, data.ml.compile	ml-framework.md

Verified against the registered tool names in alphaswarm/data/mcp/tools/ml.py (and hetegnn.py) — the tool slice does not currently expose data.ml.predictors.*, data.ml.experiments.*, or data.ml.finetune.* prefixes despite the Predictor Hub / AlphaBacktestExperiment / finetune-trainer concepts existing in alphaswarm_models.

Deployment surfaces

Surface	Where
Compose	folded into api
Kustomize	folded into base/alphaswarm-core/
AlphaSwarm CR	folded into AlphaSwarmMonolith — no dedicated toggle; scales with spec.apiReplicas since AQPMonolithSpec has no mlMcp field

Dependencies

Upstream:

• mlflow (5000) — experiment + model registry.
• postgres — Predictor Hub catalog.
• polaris / minio — feature store reads.
• bentoml / kserve (when serving backend = remote) — model invocations.

Downstream:

• Agentic plane (alphaswarm/agents/) — ML calls go through DataMCP, never direct ORM imports.
• alphaswarm-ide research copilot.

Operations

• router_complete only: any LLM call from inside the MCP registrations goes through alphaswarm/llm/providers/router.py (rule 2).
• OOD guard + circuit breaker: the MLSkillRuntime applies rules/ood_guard.py and the circuit breaker before model calls.
• Audit: every tool invocation lands an agent_runs_v2 row.

See also

• mlops-service.md — MLOps service contract.
• data-mcp.md — DataMCPTool boundary.
• mcp-rfc-conformance (private alphaswarm_internal repo) — RFC 9728 + RFC 8707 conformance. (As of 2026-07-19, alphaswarm/.cursor/rules/ only carries a pointer to this SSoT repo.)