mcp-assist-memory
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@mcp-assist-memoryRemember that the API design decision was to use REST for users."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
mcp-assist-memory
A remote MCP server (Streamable HTTP) that gives Claude a shared memory layer
across surfaces — claude.ai web, Claude Code CLI, and Claude Code Desktop —
so work state survives surface switches. It stores append-only revisioned
memory entries, work-session timelines, cross-surface handoffs, and uploaded
artifacts (with automatic ingestion of debug-capture session ZIPs), all in
SQLite plus a filesystem blob store behind a single StorageBackend
interface. It is memory-only by design: no third-party credentials, no
outbound API calls.
Full contract: see SPEC.md.
Environment variables
Variable | Required | Default | Meaning |
| yes | — | Bearer token; server refuses to start without it |
| no |
| SQLite DB + blob store location (must be persistent storage) |
| no |
| Per-upload size cap |
| no |
| Global storage cap |
| no |
| HTTP port (Replit sets this automatically) |
Related MCP server: Chronica
Run locally
pip install -e ".[dev]"
MCP_AUTH_TOKEN=dev-token python main.py
# health: curl http://localhost:8000/ → {"status":"ok"}
pytestDeploy on Replit
Import this repo into Replit. The included
.replitmakes the Run button work (python main.py).Add a Secret
MCP_AUTH_TOKENwith a long random value (e.g.python -c "import secrets; print(secrets.token_urlsafe(32))").Deploy as a Reserved VM (recommended): the server is stateful and long-running; Autoscale deployments can cold-start and run multiple instances, which breaks SQLite assumptions.
⚠️ Persistence caveat:
DATA_DIRdefaults to./datainside the workspace. The workspace filesystem persists in the editor but a deployment gets a fresh copy of the repo on each redeploy — anything written at runtime under the deployment's filesystem is lost on redeploy. PointDATA_DIRat storage that survives redeploys (e.g. a mounted persistent disk on the Reserved VM), or accept that a redeploy resets memory. Do not commitdata/to git (it's.gitignored).Your endpoint is
https://<your-repl-url>/mcp.
Register the server on each surface
Claude Code CLI / Desktop:
claude mcp add -s user --transport http assist-memory \
https://<repl-url>/mcp \
-H "Authorization: Bearer <token>"claude.ai web: Settings → Connectors → Add custom connector, with URL
https://<repl-url>/mcp. Note: the web connector UI authenticates via
OAuth and does not currently let you attach a custom Authorization
header; this server only supports static bearer auth. If your connector
form has an advanced/header field, use Authorization: Bearer <token>.
Otherwise web access requires fronting the server with an OAuth-capable
proxy — until then, use the CLI/Desktop registration and handoff_save /
handoff_load to move state to and from web sessions.
Tool overview
Group | Tools |
Memory |
|
Sessions |
|
Handoff |
|
Artifacts |
|
Meta |
|
Memory is append-only: every write is a new revision, deletes are
tombstones, and memory_revert restores by copying — history is never lost.
Uploading a debug-capture ZIP (a session.json export with
schema_version "1.0") auto-creates the session record and stores its
agent-handoff/brief.md as a queryable memory entry
(debug/<session_id>/brief).
Security
Every request to
/mcprequiresAuthorization: Bearer $MCP_AUTH_TOKEN(constant-time compare); the only anonymous route isGET /.ZIP uploads are checked for zip-slip, absolute paths, symlinks, entry count (≤ 2000), and decompression bombs (≤ 4 ×
MAX_UPLOAD_MB).Values matching common credential patterns are stored but tagged
possible-secretwith a warning in the response.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/aliomraniH/mcp-assist-memory'
If you have feedback or need assistance with the MCP directory API, please join our Discord server