A memory MCP server for coding agents
One MCP endpoint that every coding agent on your team reads and writes: decisions, fixes and conventions, typed, searchable and shared by repository.
What a memory MCP server does
The Model Context Protocol (MCP) is an open standard for connecting AI agents to tools and data. A memory MCP server gives an agent tools to store what it learns and search it later, so the next session does not start from zero. Recallium is a memory MCP server built for software teams: it keeps the record coding agents produce across the lifecycle and shares it across tools and teammates.
What to look for in a memory MCP server
| Question | Recallium |
|---|---|
| What is a memory scoped to? | A project, derived from the Git repository, and a workstream inside it |
| Does it understand software work? | Yes: it follows how engineering teams work, from design and decisions through debugging and shipping, and keeps track of which decision replaced which |
| Can a team share it? | Yes, by repository on Cloud Pro; a repository only you connect stays yours |
| How is it searched? | Hybrid search by intent and by file path; agents expand only the results they need |
| Is retrieval measured? | 99.8% hit@10 on LongMemEval-S from 3.8 results on average, protocol published |
| Which agents connect? | 60+ MCP clients; the installer wires 16+ coding agents for you |
| Does it scale? | Stress tested to 200M memories, with under 200ms average search response |
Tools the Recallium Cloud server exposes
| Tool | What the agent uses it for |
|---|---|
recallium | Loads a project’s context in one call: recent work, working state, rules, workstreams and open tasks |
search_memories | Searches by intent, by file path, or both |
expand_memories | Returns the full content of the results the agent picked |
store_memory | Stores a memory with its files and workstream |
create_workstream / get_workstream | Groups the memories of one effort and reads its decisions and designs |
set_working_state | Leaves a resume note for the next session, agent or teammate |
store_rule / get_rules | Keeps team rules that load into every session |
Connect it
npx -y recallium install detects Claude Code, Codex, Cursor, GitHub Copilot, VS Code and the other agents on your machine and adds a Recallium Cloud server named recallium to each (quickstart).
Comparing memory servers? See Recallium against CLAUDE.md, Cursor Rules, Mem0 and Supermemory.