Straight answers

cMCP is built on open standards (MCP and REST), so any AI app that speaks MCP or REST can connect. Claude Code and Claude Desktop are available now. Multi-vendor access, including ChatGPT, Gemini, Copilot, Cursor, Codex, Gemini CLI, Cline, local models and Claude on the web and mobile, is built and in testing; anything that does not pass testing will be removed. Bots connect with API keys, and your own applications can use the REST API.

No part of cMCP requires it to. In a pilot we run cMCP in your cloud, or you self-host it on your premises. Fact extraction uses a small model running on your own server, so nothing needs to leave your infrastructure.

Not by design. A client's memory is only returned while working on that client or a project linked to it, never to another client. PostgreSQL row-level security and a deny-by-default permission service back this up, and every allow and deny is audited.

No. Connectors are read-only. They read from Jira Cloud, PostgreSQL views, IMAP mailboxes (chosen folders, sender domain verified) and Git repository addresses mapped to projects (the code is not read), and feed proposed facts into an approval queue. Microsoft 365, Google Drive and GitHub connectors are built and in testing. By design cMCP never writes into outside systems.

Not today. cMCP is offered through design-partner pilots that we run in your cloud or that you self-host. A hosted option is on the roadmap.

cMCP is pilot-stage software. It has not been independently audited or penetration tested, and we do not claim any compliance certification. An independent penetration test and SOC 2 readiness work are on the roadmap. Our security page lists what is built and what is not.

It depends on scope. We agree the length, the success metrics and a review point in the scoping workshop, so you know up front what you are committing to. See the pilot page for how a pilot unfolds.

When an AI app starts a task it receives a context pack: the memory inherited for its place in the organisation, labelled as data, and limited to what the person and the app are allowed to see. Every result is permission-checked, and the audit log can show which memory versions sat behind an answer.

Personal memory is private and is never promoted. Shared scopes change only through a role, a time-limited grant or an approval, and session summaries are confirmed by the person before they are saved.

Your memory stays with you. cMCP is independent of any one AI vendor: you can change vendor, or use several at once, without losing what your organisation has learned. Multi-vendor access is built and in testing. The audit log records which model was used, so you can see the effect of a change.

cMCP is built on open-source components with permissive licences, and it exposes open interfaces: MCP and a REST API described in OpenAPI 3.1. We have not announced plans to release cMCP's own source code. If source access matters to your evaluation, tell us.

There is no price list. Pilots are scoped per organisation, so please contact us and we will talk it through.

Probably not. cMCP is designed for organisations where many people and AI agents share projects. Very small teams, companies that barely use AI, and one-off work are usually better served by something simpler.