Epic: MCP platform — shared-core decision + usage telemetry

[JSONbored]

Problem

MCP is loopover's primary agent-facing surface for both AMS (miner) and ORB (review) work, and today it's split across two independently-implemented codebases: packages/loopover-mcp/bin/loopover-mcp.js (the published, locally-installed --stdio wrapper, ~4,485 lines) and src/mcp/server.ts (the remote/hosted server, ~4,379 lines). This session's gardening pass already found and filed 6 concrete tool-coverage parity gaps between them (#6149, #6150, #6151, #6152, #6153, #6154). This epic is the layer above that: deciding whether the two should share an implementation core (not just matching tool lists), and building real usage telemetry so future prioritization is evidence-based instead of guesswork.

Area

packages/loopover-mcp/, src/mcp/server.ts, src/api/routes.ts (auth endpoints), a new telemetry sink.

Proposal

Two sub-specs, each a genuine architecture/design decision:

1. Shared-core decision — should tool implementations (not just tool names) be de-duplicated between local and remote, given packages/loopover-miner/lib/github-token-resolution.js's own header comment already explains why a hard runtime dependency between @loopover/mcp and @loopover/miner was deliberately avoided ("separately-installable CLIs... installing one doesn't require the other")? The same tension likely applies here.
2. Telemetry strategy — instrument real per-tool, per-caller-type usage tracking (which tools get called, how often, by local vs. remote callers) so future MCP prioritization stops being guesswork. Must respect this repo's existing strong privacy boundary (README: "MCP local branch analysis sends metadata, not source contents") — no source content, no PR body/diff text, no wallet/hotkey/trust-score data in any tracked event.

Auth itself is not included as a separate spec here — loopover-mcp login's GitHub App device-flow (landed this week, #6114-#6117) already appears to deliver a frictionless single-login experience for both products; if either sub-spec surfaces a remaining rough edge during design, fold it in rather than filing a third spec speculatively.

Deliverables

• Sub-spec: shared-core decision for local vs. remote MCP tool implementations (see linked sub-issue).
• Sub-spec: telemetry/usage-instrumentation strategy for MCP (see linked sub-issue).
• Tool-coverage parity work tracked separately: feat(mcp): register the 8 write-tools (open_pr/file_issue/apply_labels/etc.) that packages/loopover-mcp's own recommendedTools already promise #6149, feat(mcp): register plan-DAG tools + local scorer (build_plan/plan_status/record_step_result/run_local_scorer/predict_gate) in packages/loopover-mcp #6150, feat(mcp): register 4 maintainer/repo-owner intake tools in packages/loopover-mcp (get_skipped_pr_audit/get_issue_quality/get_registration_readiness/get_config_recommendation) #6151, feat(mcp): register loopover-mcp maintain queue/approve/reject/pause/resume/set-level/precision as MCP tools (CLI-only today) #6152, fix(mcp): packages/loopover-mcp's MAINTAIN_AUTONOMY_LEVELS still accepts "suggest"/"propose", values the server rejected since #4620 #6153, feat(mcp): add an MCP tool for GET /v1/repos/:owner/:repo/pulls/:number/reviewability (route already MCP-auth-aware, no tool on either server) #6154 (already filed, not duplicated here).

Resources

• packages/loopover-mcp/bin/loopover-mcp.js, src/mcp/server.ts
• packages/loopover-miner/lib/github-token-resolution.js (the existing "separately-installable CLIs" rationale to weigh against)
• #6114-#6117 (the GitHub App device-flow auth work this epic builds on, not duplicates)

Boundaries

• maintainer-only until both sub-specs land a real decision — this is architecture and privacy-boundary judgment, not mechanical engineering.
• No telemetry event may carry source code, PR content, wallet/hotkey material, or trust/reward scores — matches this repo's existing privacy commitments.
• Neither sub-spec should assume the other's outcome; a shared-core decision doesn't require a particular telemetry answer and vice versa.

[linear-code]

JSO-4314

[JSONbored]

Both sub-specs now resolved and closed: #6227 (shared-core decision) and #6228 (telemetry strategy, just closed — fully shipped via #6235-#6238). All 5 sub-issues complete.

[JSONbored]

All 5 sub-issues complete (#6227, #6228, #6245, #6300, #6301). Closing this epic.