systems

Compatibility is about the client, not the model. If a host supports MCP, any model it supports can use radical memory.

supported now

supported
last updated: 21 Dec 2025
claude desktop / claude code
Native MCP support. Works great as a first-class memory tool.
live
cursor
MCP tools support. Ideal for builder workflows.
live
cline
Works via MCP tools and resources.
live
continue
MCP support plus flexible model configuration.
live
chatgpt (connectors)
Remote MCP servers can surface through supported connector paths.
live
openai codex (cli and ide)
MCP-enabled tooling can call radical memory inside codex workflows.
live
gemini via mcp-capable clients
Works wherever Gemini is used through an MCP client path, including CLI style flows.
live

core capabilities

A small, stable MCP surface area. Same behavior across every supported client.

Search memory
Fast retrieval for the current moment. `memory.search(query, project_id, limit)`
Context packs
A token-budgeted context pack you can inspect and trust. `memory.get_context_pack(project_id, user_message, budget_tokens)`
Write memory
Store stable facts: decisions, preferences, project state. `memory.store(type, content, project_id, source)`
Pin + forget
Keep what matters. Remove what doesn't. `memory.pin(memory_id)` and `memory.forget(memory_id)`

coming soon

coming soon
more client integrations
Broader coverage across the MCP client ecosystem as it grows.
soon
context pack viewer
See what memory was pulled in, and why, before it reaches a model.
soon
team workspaces
Shared memory with permissions, audit history, and portable context packs.
soon
policy engine
Rules for what gets stored, redacted, pinned, or expires automatically.
soon