Provena Agent Memory — Claude Code MCP server: install and setup
Provena Agent Memory maintains cross-session agent context linked directly to raw evidence. Helpful when an assistant needs verifiable memory across work runs.
MIT ★ 8 updated 2026-10-19
claude mcp add --env PROVENA_API_URL=<PROVENA_API_URL> --env PROVENA_API_KEY=<PROVENA_API_KEY> --env PROVENA_SCOPE_ID=<PROVENA_SCOPE_ID> --transport stdio provena-agent-memory -- uvx provena-agent-memoryWhat Provena Agent Memory does
The server links each stored memory to the exact interaction where it originated. Instead of keeping isolated notes, it stores the source event, review stage, and tenant scope. The memory_explain tool lets developers inspect why a claim exists and whether earlier statements contradicted it.
Under the hood runs a FastAPI backend with PostgreSQL and pgvector, backed by Ollama or hosted inference. A Next.js web dashboard provides an interface to review extracted claims manually. All events and evidence remain strictly append-only.
Running the full stack locally demands Python 3.11+, pipx, and Docker Compose. Initial setup downloads approximately 1.26 GB of model weights for Ollama. Scopes require an exact identifier match, so branch context inheritance does not work yet.
What it helps with
- trace agent facts back to originating dialogue turns
- review candidate claims through a dedicated web dashboard
- isolate memory records by project or tenant scope
- audit how preferences and assertions evolve over time
Example prompts for the agent
- Run
memory_explainto show where you learned the deployment targets for this service. - Check your stored memory records for this scope before updating the architecture.
How to install Provena Agent Memory
Platform CLI
claude mcp add --env PROVENA_API_URL=<PROVENA_API_URL> --env PROVENA_API_KEY=<PROVENA_API_KEY> --env PROVENA_SCOPE_ID=<PROVENA_SCOPE_ID> --transport stdio provena-agent-memory -- uvx provena-agent-memoryReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from provena-agent-memory <command>
Platform CLI
codex mcp add provena-agent-memory --env PROVENA_API_URL=<PROVENA_API_URL> --env PROVENA_API_KEY=<PROVENA_API_KEY> --env PROVENA_SCOPE_ID=<PROVENA_SCOPE_ID> -- uvx provena-agent-memoryReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from provena-agent-memory <command>
Config file
{
"mcpServers": {
"provena-agent-memory": {
"command": "uvx",
"args": [
"provena-agent-memory"
],
"env": {
"PROVENA_API_URL": "<PROVENA_API_URL>",
"PROVENA_API_KEY": "<PROVENA_API_KEY>",
"PROVENA_SCOPE_ID": "<PROVENA_SCOPE_ID>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from provena-agent-memory <command>
Config file
{
"servers": {
"provena-agent-memory": {
"type": "stdio",
"command": "uvx",
"args": [
"provena-agent-memory"
],
"env": {
"PROVENA_API_URL": "<PROVENA_API_URL>",
"PROVENA_API_KEY": "<PROVENA_API_KEY>",
"PROVENA_SCOPE_ID": "<PROVENA_SCOPE_ID>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from provena-agent-memory <command>
Platform CLI
gemini mcp add -e PROVENA_API_URL=<PROVENA_API_URL> -e PROVENA_API_KEY=<PROVENA_API_KEY> -e PROVENA_SCOPE_ID=<PROVENA_SCOPE_ID> provena-agent-memory uvx provena-agent-memoryReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from provena-agent-memory <command>
Step 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 Provena Agent Memory
How to install Provena Agent Memory in Claude Code?
Run claude mcp add provena-memory -e PROVENA_API_KEY=<PROVENA_API_KEY> -e PROVENA_SCOPE_ID=<PROVENA_SCOPE_ID> -- uvx provena-agent-memory in your terminal. Replace the placeholders with your actual API key and scope identifier.
What does Provena Agent Memory do?
It acts as an MCP memory backend that attaches raw conversation proof to agent recollections. You can see the original quote behind every remembered statement.
What dependencies are needed to host Provena locally?
You need Docker Compose, Python 3.11+, and pipx. The initial quickstart also pulls roughly 1.26 GB of model data for the CPU Ollama container.