You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I searched existing issues and docs before opening this.
This proposal is grounded in a real Roomote use case, not a broad generated feature list.
Problem
Hi I am Vivek Gupta, Founder & CEO at memcode.in
Roo Code coordinates a team of AI agents inside the code editor. MemCode could preserve repository context, engineering decisions, user preferences, and agent handoffs across coding sessions so the team does not repeatedly rediscover the same project knowledge.
Current workaround
Memory is not present
Proposed behavior
MemCode can provide this as an optional integration rather than a rewrite: an MCP server for users running agent workflows, SDK/API hooks for native persistence, and a self-hosted/local or enterprise deployment where data must stay in the operator’s environment. We can keep it opt-in and scoped to saved memories such as decisions, preferences, project facts, and approved context.
Affected surfaces
Generalist agent
Self-hosting implications
No, If this fits your roadmap, I can open a focused PR or do a short call with maintainers. If you turn it into a paid integration or managed add-on, we’re open to discussing a 10% revenue share on attributable MemCode-related revenue, subject to your contributor, legal, and commercial policies, no commitment is required to evaluate the technical fit. https://memcode.in/leviathan
Alternatives considered
No response
Documentation impact
No response
cc @daniel-lxs for your thoughts on this integration.
Thanks for the proposal. Roomote already has opt-in durable Brain memory, a provider-neutral memory MCP path used by Supermemory (category: 'memory'), and support for custom remote or local MCP servers. Based on the current architecture, the likely fit would be another provider/catalog integration rather than a new native persistence subsystem or SDK hook.
I need some clarification before a concrete implementation plan can be grounded in the repository. The linked Leviathan page does not currently expose these protocol details:
What is the canonical hosted Streamable HTTP MCP endpoint, and is there public MCP/tool documentation?
Which exact tools and input/output schemas does MemCode expose for recall/search, create/update, and delete/forget operations?
What authentication contract does the hosted endpoint use: MCP OAuth discovery with PKCE/Dynamic Client Registration, a pre-registered OAuth client, or header/API-key authentication? Are any scopes, plan restrictions, beta access, or allowlisting required?
How are deployments, members, repositories, and environments isolated so context cannot be recalled across the wrong tenant or project? Which stable identifiers must the client provide?
Does Leviathan expose the same endpoint, transport, auth behavior, and tool schemas as the hosted service? What configuration and networking assumptions apply to self-hosted installations?
What provenance, retention, export, and explicit delete/forget guarantees are available, and can operators verify or constrain what was persisted?
Is there a test instance or reproducible setup maintainers can use to validate task-start recall, completion-time writes, private-task behavior, and multi-provider coexistence?
Once that contract is available, maintainers can evaluate whether a built-in catalog entry adds enough value beyond Roomote's existing custom MCP path and, if so, scope the catalog metadata, connection flow, runtime tests, UI branding, and public documentation.
Please tag @roomote in your response with the details, and I can turn them into a concrete repository-grounded implementation plan.
Activity
roomote-community commentedon Sep 21, 2026
Thanks for the proposal. Roomote already has opt-in durable Brain memory, a provider-neutral memory MCP path used by Supermemory (
category: 'memory'), and support for custom remote or local MCP servers. Based on the current architecture, the likely fit would be another provider/catalog integration rather than a new native persistence subsystem or SDK hook.I need some clarification before a concrete implementation plan can be grounded in the repository. The linked Leviathan page does not currently expose these protocol details:
Once that contract is available, maintainers can evaluate whether a built-in catalog entry adds enough value beyond Roomote's existing custom MCP path and, if so, scope the catalog metadata, connection flow, runtime tests, UI branding, and public documentation.
Please tag @roomote in your response with the details, and I can turn them into a concrete repository-grounded implementation plan.