Roadmap H2 2026 — protocol readiness, reliable writes, and measurable adoption

[tuanle96]

Purpose

This roadmap supersedes #27. The original product thesis remains: odoo-mcp is the free, zero-Odoo-side-setup, bring-your-own-LLM layer for Odoo 16+. The tracking model changes from version-shaped wish lists to outcomes with explicit owners, dependencies, acceptance gates, and proof.

Current released baseline: v1.2.1. Most of #27's v1.1–v1.3 scope actually shipped in v1.1.0, so release numbers are no longer used as the primary roadmap hierarchy.

Status vocabulary

• planned — scoped, not started
• active — implementation or review is moving
• external-wait — blocked on an upstream maintainer/spec/SDK
• shipped — present in a published release with proof
• dropped — explicitly removed with rationale

Lane A — Core MIT / self-hosted

A0. Clear the contributor flight deck — active

• Re-review and merge feat(schemas): add typed output schemas for remaining tool groups #50 when runtime payload validation, mypy, structural/architecture gates, and CI are green; merged as 8b40df72, and Typed output schemas: knowledge + accounting + async + cross-instance (12) #30 is closed.
• Re-review and merge refactor: extract addon source scanner #51 only if the legacy agent_tools monkeypatch surface still behaves end-to-end; current head still fails this behavioral contract and has a maintainer CHANGES_REQUESTED review.
• Bring draft Add typed output schemas for write tools #40 back to current main, require runtime-payload proof, then close Typed output schemas: write tools (5) #28 when shipped.
• Split or assign Typed output schemas: diagnostics tools (11) #29 in reviewable batches rather than letting an 11-tool schema sweep sit unowned.
• Field-ranking extraction shipped via Extract field ranking helpers into dedicated module #42; close Split agent_tools.py: extract field ranking + smart field selection #32 as delivered in v1.2.1.

Exit gate: no near-complete PR is held by stale review state, and each contributor issue links to its current implementation proof.

A1. MCP 2026-07-28 compatibility — planned

Tracked in #52. Ship as a dedicated compatibility release after the final spec and stable Python SDK v2 are available.

Exit gate: protocol conformance, client/transport matrix, auth regression proof, full quality gates, migration notes, and rollback path.

A2. Complete public output contracts — active

• Read tools: shipped in v1.1.0.
• Knowledge/accounting/async/cross-instance: merged to main via feat(schemas): add typed output schemas for remaining tool groups #50; awaiting the next batch release before status becomes shipped.
• Write tools: Typed output schemas: write tools (5) #28 / Add typed output schemas for write tools #40.
• Diagnostics: Typed output schemas: diagnostics tools (11) #29.

Exit gate: every public tool schema is validated through the actual MCP runtime conversion path for representative success and error payloads; CI and mypy are green.

A3. Durable write approvals across workers — planned

Tracked in #49. The preferred direction is an atomic shared WriteApprovalStore, preserving consume-once semantics, server-only binary staging, instance/principal binding, and fail-closed behavior.

Exit gate: validate on worker A / execute on B, concurrent claim-once, expiry, backend outage, restart/eviction, _from_path, crash/unknown-outcome, and session-affinity distinction are proven. Security/advisor review is mandatory.

A4. Migration evidence gate — planned

Tracked in #53. The shipped taxonomy remains canonical; the missing work is the reproducible artifact and interpretation contract.

Exit gate: schema, clean/blocked fixtures, coverage caveats, regression tests, and consultant-facing interpretation guide.

Lane B — Adoption and ecosystem

Distribution

• Glama: shipped and installable.
• GitHub release .mcpb: shipped and attached automatically.
• Docker MCP Catalog: external-wait on feat: add odoo-mcp MCP server docker/mcp-registry#3928.
• Official MCP Registry workflow: shipped repo-side; verify each release publication.
• PulseMCP and mcp.so: planned; require live listing proof before marking shipped.

Measures that matter

• time to first successful health_check;
• setup failure categories by client;
• .mcpb install success rate;
• usage of data-quality and migration workflows;
• externally maintained plugins passing a compatibility suite;
• issue/PR first-response time and release-scope drift.

Operating rule: release monthly at most, except urgent fixes. Every release must include a scope statement and a roadmap delta.

Lane C — Cloud/control plane after Gate 0

The core remains MIT and self-hosted. A paid control plane may own tenant/runtime management, per-user OAuth/SSO, encrypted credentials, centralized audit UI, and fleet reports—but build-out waits for all three demand signals:

• 3 paid audit customers
• 10 qualified waitlist emails
• 5 agency discovery conversations

Until then, sell manual audit/services and collect Voice of Customer evidence. Do not move cloud-only infrastructure into the community core roadmap.

Maintainer invariants

• First response target: under 48 hours, measured rather than asserted.
• Contributor credit in every applicable CHANGELOG/release entry.
• No feature is marked shipped without release/evidence links.
• Active items are reviewed at least every 14 days; stale items are reclassified.
• Public API, auth/trust-boundary, and cross-layer completion must pass the mandatory advisor protocol.

Baseline proof

• v1.1.0 shipped OAuth hardening, server instructions, typed read schemas, data quality, migration workbench, plugins, tool filtering, and .mcpb.
• v1.2.0/v1.2.1 shipped a different release train, which is why Roadmap v1.x — the AI layer of the Odoo community #27 is being superseded rather than silently rewritten.
• Read-only roadmap assessment: plans/reports/assessment-260716-roadmap-issue-27.html in the maintainer checkout.

[tuanle96]

Roadmap delta — 2026-07-29

• Baseline is now v1.3.0, published with MCP 2026-07-28 and stable Python SDK v2; MCP 2026-07-28 + Python SDK v2 compatibility release #52 is completed.
• refactor: extract addon source scanner #51 has been re-reviewed against current main and approved: temporary merge proof is 917 tests, Ruff, mypy, and 2/2 import contracts. Close Split agent_tools.py (1,317 LOC): extract the AST addon scanner #31 after merge.
• Draft Add typed output schemas for write tools #40 was closed as stale/superseded; Typed output schemas: write tools (5) #28 remains open for a fresh SDK-v2-native write-schema PR.
• fix: fence uncertain approved write replays #61 now has an atomic session-local approval claim and concurrent regression on its latest head; this remains distinct from the multi-worker shared-store scope in Write-approval durability across multiple worker processes (gate 2: process-local store) #49.
• feat: route long single-field payloads through the local filesystem #55 remains changes-requested and needs a current-main rebase plus a focused field-file I/O diff.
• PyPI, GHCR, Smithery, the official MCP Registry, GitHub Release, and .mcpb are live for v1.3.0.

The issue body should be refreshed from the old v1.2.1/planned baseline at the next roadmap edit.