MCP server for 0Latency — persistent memory layer for AI agents. Give Claude, Cursor, or any MCP client long-term memory.
Free
Nothing to run from here
Engrams is installed from its publisher's own source and answers where it runs, so this marketplace is not in the path of a single call. There is no address here to send one to, and a panel that pretended otherwise would be showing you an answer we made up. Install it and call it from your own client — the Installation tab has the entry for each one.
Open Installation—
p95 latency
—
call success
local
runs on your machine
—
calls last 7d
What it does
Universal AI memory layer — MCP server with persistent, cross-tool memory for Claude Code, Cursor, Windsurf, and more
Quickstart
# 1 — run it from where its publisher ships it
npx -y engrams
# 2 — the plan is chosen at checkout, in the browser
# 3 — ask your agent something
> Universal AI memory layer — MCP server with persistent, cross-tool memory for Claude Code, Cursor, Windsurf, and more
Publishernot claimed
SUAll 1 server
sunrise-labs-dot-ai1 server
1 category
Collected from a public index. Nobody has claimed this account, so nothing here was written by its author — claim it if it is yours.
Installation
Where are you running it?
Every route below installs the same thing and ends at the same approval screen. This one runs on your machine: your client starts Engrams as a process under your own user, with your files and your network, so the tool surface below is what it can do to your computer rather than to a server somewhere else. It is scanned, signed and pinned to the version you choose — read the surface before you approve it.
Claude Codedeclared by this listing
This is a public server: you run it yourself and this marketplace is not in the path. Claude Code registers it in one command.
claude mcp add engrams-mcp -- npx -y engramsReconnect, or start a new session, and the tools appear in the model’s tool list.
This is a paid listing, so the CLI has to be signed in and the licence has to be on your account — how the login works. Until then the install stops with the price rather than half-writing a config entry.
What it writes
One config entry your client uses to start the process locally. A local server runs with your file system and your network, which is why it is priced without metering.
What you approve
40 tools, with what each one reads, writes and reaches shown before you agree — the same list on every route above. Read the tool surface.
Tool surface
What the model actually sees. Descriptions are diffed on every release — see version history.
Store a new memory. If a similar memory already exists, returns the existing memory and resolution options instead of writing immediately. Call again with
Takes no parameters.
Example callreturns →
Upload many memories at once, bypassing dedup by default. Designed for imports from canonical external sources (Google Contacts, ChatGPT export, a contacts CSV) where the caller has already deduped against its own source. Each entry
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Check the staleness of indexed documents. Returns which documents haven
Takes no parameters.
Example callreturns →
Search the user
Takes no parameters.
Example callreturns →
Token-budget-aware context search optimized for LLM consumption. Returns memories packed into a specified token budget, organized hierarchically or as a narrative block.\n\n
Takes no parameters.
Example callreturns →
Report which memories from a prior memory_context call you actually cited vs. filtered as noise. Call this BEFORE ending your turn — unrated retrievals are wasted learning signal. Pass the
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Delete a memory the user no longer wants stored. Call when the user explicitly asks to forget something.
Takes no parameters.
Example callreturns →
Soft-delete many memories at once, scoped by a filter. Use to retire benchmark / test data, drop a whole domain, or de-pollute by entity name. ALWAYS dry-runs first by default — pass dryRun=false explicitly to commit. Filter requires at least one of
Takes no parameters.
Example callreturns →
Confirm a memory is still accurate. Call this when you act on a memory and the user validates it was correct, or when the user reaffirms something you already know.
Takes no parameters.
Example callreturns →
Replace a memory
Takes no parameters.
Example callreturns →
Flag a memory as wrong or outdated. Call this when you act on a memory and the outcome was incorrect, or when the user says a memory is no longer true. Degrades confidence by 0.15.
Takes no parameters.
Example callreturns →
Create a relationship between two memories. When
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Split a memory that contains multiple distinct facts into separate memories. Call this when a memory covers more than one topic or would be more useful as individual pieces. Each new memory inherits the original
Takes no parameters.
Example callreturns →
Scan a memory for sensitive data (PII, API keys, etc.) and optionally redact it. Use this to check memories before sharing or to clean up accidentally stored secrets.
Takes no parameters.
Example callreturns →
Get all connections for a memory
Takes no parameters.
Example callreturns →
List all memory domains with counts. Returns per-domain metadata: count of memories, whether it
Takes no parameters.
Example callreturns →
Board advisory work
Takes no parameters.
Example callreturns →
Archive a life domain. Active snippets are preserved. Future memory_write_snippet calls to this domain are rejected until it is unarchived via memory_register_domain. Requires write permission on the target domain. Returns status=
Takes no parameters.
Example callreturns →
Browse the user
Takes no parameters.
Example callreturns →
Fetch one or more memories by ID (returns full record contents — including any PII). Use when you have a memory ID from a prior search/context call, a deeplink, or another tool
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Export memories as JSON for migration to another Lodis instance. Returns paginated results — call repeatedly with increasing offset until hasMore is false. Use with memory_import (source_type:
Takes no parameters.
Example callreturns →
List unclassified memories that need entity_type assignment. Returns memories without entity types so you can classify them using memory_update. Use this to find memories that need classification.
Takes no parameters.
Example callreturns →
List all known entities grouped by type. Useful for discovering what the system knows about people, organizations, projects, etc.
Takes no parameters.
Example callreturns →
Set per-agent read/write permissions for a domain
Takes no parameters.
Example callreturns →
Pin a memory as canonical — permanent knowledge that is immune to confidence decay. Use for confirmed facts, preferences, skills, or lessons that should never fade.
Takes no parameters.
Example callreturns →
Archive a memory — preserves it for reference but deprioritizes it in search results. Confidence is frozen. Use for completed project context or outdated but historically relevant information.
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Takes no parameters.
Example callreturns →
Migrate memories between local and cloud storage. Use
Takes no parameters.
Example callreturns →
Interactive tutorial for learning how Lodis works. Call with no args for a table of contents, or with a chapter id (e.g.
Takes no parameters.
Example callreturns →
Standardmost installed
$15 / month
- 12,000 calls a month, included
- 144 calls a minute at peak
- Every tool the listing has
- Cancel any time — the unused days are credited
- Runs behind the gateway, so cancelling stops it
Scale
$59 / month
- 100,000 calls a month, included
- 1,200 calls a minute at peak
- Every tool the listing has
- Cancel any time — the unused days are credited
- Runs behind the gateway, so cancelling stops it
What counts against your monthly calls
| Tool | Unit | Calls used | Out of the allowance |
|---|
No call is billed on its own, and this plan carries no monthly call limit. $15 a month covers every tool at any volume.
Call allowance · this installenforced at the gateway
Monthly allowance, in calls
—
—
Alert at 80% · then the tool returns a normal MCP error
Used this month: $8.42
$16.58 left
ATrust score 99 / 100
Release 0.5.1 · scanned 25 Aug 2026
Two independent axes, because powerful and malicious are different questions. The grade is threat only. The capability level is blast radius, and it is never a penalty on the grade — it is priced as one subtract-only term in the score, where you can see it.
Threat score100
Client score99
Findings0 threat · 0 capability
Capabilityminimal
Coveragesource
Verificationrepo
Surface digestc558dccd160fd226ddc0a7de40af6895b4ccee46585f7a3dfafb17af0db05e5d
Score vectorthreat 100 → client 99
| Term | Level | What it prices | Points |
|---|---|---|---|
| capability-exposure | minimal | capability blast radius (minimal) — client exposure if the model is manipulated | −0 |
| verification-discount | repo | publisher verification (public source) — no provenance, but the source is public and inspectable | −1 |
| coverage-honesty | source | inspection depth (source) — how much of the target the scan could see | −0 |
score = clamp(round(threatScore − E_cap − E_ver − E_cov), 0, 100). No model, no opinion, no network call at scoring time: the same methodology version against the same bytes always produces this number, and every line above is a field of the report rather than a summary of it.
What the scan could actually read
A grade is only as meaningful as its coverage, so the scanner publishes its own depth before it publishes its result.
Tool surfaceread
Implementation sourceread
Package metadataread
Live transportnot available
Tools were statically extracted from the published source (27 recovered), not enumerated from a running server. Tool-poisoning, Unicode-smuggling, capability and toxic-flow analysis ran on this inferred surface, but a mis-parsed registration could be missed or mis-attributed, so tool-derived findings are capped below “confirmed”. To grade the real runtime surface, scan the running server: --command "npx -y <package>".
Capability — what it could do if the model were manipulated
Tags derived from each tool’s schema and the implementation, not from what the tool calls itself. minimal is the level these add up to.
| Tool | Capability tags | Why the tag was assigned |
|---|---|---|
| memory_write | no tags | |
| memory_index | no tags | |
| memory_index_status | no tags | |
| memory_search | no tags | |
| memory_context | no tags | |
| memory_briefing | no tags | |
| memory_update | no tags | |
| memory_remove | no tags | |
| memory_confirm | no tags | |
| memory_correct | no tags | |
| memory_flag_mistake | no tags | |
| memory_connect | no tags | |
| memory_split | no tags | |
| memory_scrub | no tags | |
| memory_get_connections | no tags | |
| memory_list_domains | no tags | |
| memory_list | no tags | |
| memory_export | no tags | |
| memory_classify | no tags | |
| memory_list_entities | no tags | |
| memory_set_permissions | no tags | |
| memory_pin | no tags | |
| memory_archive | no tags | |
| memory_onboard | no tags | |
| memory_interview | no tags | |
| memory_import | no tags | |
| memory_migrate | no tags |
Toxic-flow graph
The lethal trifecta, checked as a graph rather than as a checklist: untrusted input, a sensitive source and an external sink have to meet before there is a path worth worrying about.
The public result for this release does not print the flow graph, so there is nothing to show here. That is not the same as "no paths were found": what the scan did read is above, under coverage.
Supply chain and provenance
Integrityfirst-seen
Surface digestc558dccd160fd226
Scanned2026-08-25
This is the first scan of this surface here, so there is nothing yet to compare it against.
Every result on this tab comes from one deterministic pass over the published package — offline, rule by rule, and auditable line by line above. Same methodology version, same bytes, same score.
Release history
Pinned to 0.5.1 — the install command below asks for that release. A pin is part of an install, so it is kept for this visit and written down when you install.
0.5.1
undated
installedNo release note was published with this version.
—
no reviews yet
Reviews are earned
Only accounts with at least 50 real tool calls against this server in the last 90 days can post. Ratings are weighted by how much the reviewer actually uses it, and publishers can reply once per review.
Reviews are earned
Writing one takes an account with at least 50 real tool calls against Engrams in the last 90 days. That is the whole gate — there is no other way to post, which is why the counts beside each review are worth reading.
No reviews yet
Nobody has reviewed this listing. The rating on the card is the mean of the reviews written here and nothing else, so there is no rating until somebody writes the first — which takes an account with 50 real tool calls against it.