mcpscraper-memory
v1.22.0
Published
mcp-memory — a hosted, multi-tenant, access-controlled per-user Obsidian/Neon memory exposed AS an MCP server, with a scheduled optimization layer. Its own standalone MCP product.
Readme
mcp-memory
A hosted, multi-tenant MCP server for structured personal/business memory — notes across 15 purpose-built vaults, semantic search, typeahead, and real structured data tables (typed columns, filter, sort) — plus Scheduled Actions: freeform-text automations that run on a cadence, using your MCP Scraper tools to gather live data and mcp-memory to store it.
Customer setup uses one endpoint, https://mcpscraper.dev/mcp, and one long-lived credential,
MCP_SCRAPER_API_KEY, for scraping and Memory. This package is a compatibility stdio proxy to that
unified server; nothing runs or is stored on your machine.
This repository also owns the independently publishable mcpscraper-memory-db workspace for
Postgres/PgVector storage, vault contracts, Jina embeddings, and reranking, and the
mcpscraper-memory-tools workspace for public tool behavior. The hosted backend consumes their
released package versions. Release those dependencies first, then update and redeploy this service
before downstream products advertise a newer memory-tool contract.
Repository development
Run npm install once at the repository root. The root workspace then owns the server, CLI, database
package, and tool package dependency graph:
npm run build
npm run typecheck
npm test
npm run verify:workspaces
npm run verify:packagesPackage identities and release versions remain independent. See packages/README.md
for ownership boundaries and import provenance.
Connect
OAuth remote connector (Claude.ai web, Claude Desktop's "Add custom connector") — add
https://mcpscraper.dev/mcp and log in.
This package, for clients that only support local stdio MCP servers (Claude Code, Cursor, Windsurf, and similar):
{
"mcpServers": {
"mcp-memory": {
"command": "npx",
"args": ["-y", "mcpscraper-memory"],
"env": { "MCP_SCRAPER_API_KEY": "sk_..." }
}
}
}Create or rotate the same key used by MCP Scraper at https://mcpscraper.dev/dashboard.
Scheduled Actions access and billing
Scheduled Actions requires an active MCP Scraper Starter plan or higher. It is included with that plan rather than sold as a separate subscription, and there is no separate monthly run quota.
- Every started scheduled occurrence costs 75 MCP Scraper Credits.
- Agent-mode runs also add 1.5 times OpenRouter's actual reported model cost.
- Deterministic
connection_syncruns do not use a model, so they cost only the 75-Credit base. - Creating, listing, pausing, or deleting definitions does not itself consume a scheduled-run charge.
MCP Scraper is the source of truth for paid-plan access and the Credit ledger. mcp-memory retains the schedule definitions, default timezone, hosted-link key, and encrypted delegated MCP Scraper key; historical entitlement/quota columns are migration data, not current runtime gates. See the live Scheduled Actions guide and pricing.
Scheduled results and artifact templates
New scheduled executions have a durable result record separate from their automation definition
and scheduler receipt. Use list_scheduled_runs for the unopened inbox or broader history and pass
the returned opaque runId unchanged to get_scheduled_run and state-mutation tools.
list_artifact_templates returns the built-in Editorial Reading Room preset and saved templates.
Saving creates version 1; every edit creates a new immutable version. Automations select either
{"mode":"none"} or one exact saved template version. No artifact disables only rich report
rendering—it does not disable Memory-note or vault output.
The MCP resource memory://scheduled-results-contract provides the concise machine-readable
contract. Templates and rendered HTML are not embedded; existing Memory notes continue through the
normal indexing path. Expiring run-view links are issued by MCP Scraper and are separate from broad
Scheduled Actions management links.
Environment variables
MCP_SCRAPER_API_KEY— the one customer key used for scraping and Memory.MCP_SCRAPER_MCP_URL— optional endpoint override. Defaults tohttps://mcpscraper.dev/mcp.
Run npx mcpscraper-memory --help for setup instructions from the terminal.
