MemCell — Claude Code MCP server: what it does, how to install
MemCell brings persistent memory, graph traversal, and document search to AI agents. It lets models check previous context before acting and score confidence.
Apache-2.0 ★ 1 updated 2026-10-28
claude mcp add memcell -- npx -y memcellWhat MemCell does
AI coding sessions drop context once a prompt thread finishes. MemCell fixes this limitation by exposing a structured memory layer over MCP. It supplies seven workspace tools: recall for fetching instructions, remember for saving facts, report and feedback for confidence adjustments, plus document ingestion, retrieval, and telemetry insights.
Document retrieval handles PDF, DOCX, CSV, HTML, and Markdown files. Knowledge graph queries surface one-hop connections alongside text chunks. To evaluate memory reliability, the system calculates Brier scores and calibration curves.
Setup takes one command. For offline tasks without network access, running memcell start --daemon spins up an embedded PGlite vector engine locally on port 3000.
What it helps with
- retain user choices and decisions across separate agent sessions
- retrieve past preferences before executing terminal commands
- index local PDF and Markdown docs into grounded search chunks
- track outcome feedback to adjust model confidence metrics
Example prompts for the agent
- Recall our previous decisions regarding database migrations.
- Remember that unit tests must always run with the --coverage flag.
- Ingest the project design doc specification.pdf and search for auth specs.
How to install MemCell
1Platform CLI
claude mcp add memcell -- npx -y memcell2Platform CLI: remote server
claude mcp add --transport http memcell https://memcell.ai/mcp1Platform CLI: remote server
codex mcp add memcell --url https://memcell.ai/mcp2Platform CLI
codex mcp add memcell -- npx -y memcellConfig file
{
"mcpServers": {
"memcell": {
"url": "https://memcell.ai/mcp"
}
}
}{
"mcpServers": {
"memcell": {
"command": "npx",
"args": [
"-y",
"memcell"
]
}
}
}Config file
{
"servers": {
"memcell": {
"type": "http",
"url": "https://memcell.ai/mcp"
}
}
}{
"servers": {
"memcell": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"memcell"
]
}
}
}1Platform CLI: remote server
gemini mcp add -t http memcell https://memcell.ai/mcp2Platform CLI
gemini mcp add memcell npx -y memcellStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about MemCell
How to install MemCell in Claude Code?
Run claude mcp add memcell -- npx -y memcell for the local stdio setup. For a remote connection, run claude mcp add --transport http memcell https://memcell.ai/mcp.
What does MemCell do?
It connects AI coding assistants to a shared persistent memory layer. Agents query past instructions, record outcomes, and search ingested workspace documents.
Can MemCell operate completely offline?
Yes. When using the embedded PGlite backend, launching memcell start --daemon provides local vector storage without internet dependencies.