What connects
Integrations
Two connectors and no per-vendor code: a read-only SQL query, or an MCP server. Everything below is one of those two, pre-filled.
The two connectors
Adding a system of record is configuration, not an integration somebody builds for your logo.
A read-only SQL query
Postgres, Supabase or Neon. A connection string and two queries: one that lists what becomes memory, one that re-reads a row by its id. The session opens read-only.
Any MCP server
A Streamable HTTP endpoint, a tool that lists records, and a tool that fetches one. Contextely connects as the client.
Connected and checked
A system appears here once somebody has run the whole path against it: connect, sync, retrieve, watch an object pass its TTL and re-fetch. Not once the code exists.
Postgres
checked end to endOne table or view becomes memory, re-read row by row when a memory object passes its TTL.
You need: A read-only connection string, and the names of the columns that hold each row's identifier, title and text.
GitHub
checked end to endEvery issue in one repository becomes memory, and stays current as issues are edited.
You need: A personal access token with repository read access, and the owner and name of one repository.
Built, not yet checked
Selectable in the product, written against each service's published tool list, none run end to end against a real account. So none is claimed as supported.
Linear
tokenOne Linear team's issues become memory, re-read when they go stale.
Linear also offers a browser sign-in flow. Contextely uses the API key path, which is the one that keeps working unattended when a memory object goes stale at three in the morning.
Linear documentationLedgeur
tokenMeeting notes and transcripts become memory, so an agent can answer from what was actually said.
A meeting does not change after it happens, so a long TTL costs nothing here. The notes attached to one can, which is why the TTL is not infinite.
Ledgeur documentationSentry
browser sign-inYour Sentry issues become memory, across every organization the sign-in covers, so an agent can answer what is currently breaking.
Covers every organization this sign-in can see, re-read on each sync so a new one needs no reconnect. Narrowing to one organization is offered and changes only what is condensed into memory automatically: an agent can ask about anything the sign-in covers either way.
Sentry documentationGoogle Search Console & Ads
browser sign-inEvery site this Google account can see in Search Console becomes memory, each carrying its sitemap coverage. The same connection also runs Google Ads Keyword Planner searches (search volume, competition, CPC) -- enable google_ads_keyword_ideas / google_ads_keyword_metrics as actions once you have discovered this connection's tools.
Search performance figures are not stored as memory: they are a report computed per question, and a stored copy would be wrong within the hour. Enable the query and Keyword Planner tools as actions to ask for them live.
Google Search Console & Ads documentationGoogle Analytics
browser sign-inEvery GA4 property this Google account can read becomes memory, across all of its accounts, with each property's data streams and measurement ids one re-read away.
Traffic and engagement numbers are not stored as memory: they are a report computed per question. Enable the report tools as actions to ask for them live.
Google Analytics documentationNotion
browser sign-inThe pages matching one saved search become memory, re-read from Notion when they go stale.
Notion's server has no way to list everything, only to search, so a source covers one saved search rather than a whole workspace. Connect several if you want several areas.
Notion documentationSlack
browser sign-inOne channel's messages become memory, treated as a record of decisions rather than a conversation.
Slack's server requires workspace admin approval before anything can connect, and it does not support automatic client registration, so a self-hosted Contextely needs its own Slack app.
Slack documentationGoogle Drive
browser sign-inThe documents matching one saved search become memory, with the full text one call away.
Google's Drive MCP server is a developer preview, so it can change without notice. It also does not support automatic client registration, so a self-hosted Contextely needs its own Google OAuth application.
Google Drive documentation
Bring your own
Neither list is a limit. Anything publishing an MCP server is reachable now, as is any Postgres-compatible database.
Any MCP server
Anything that speaks MCP over Streamable HTTP becomes memory, with no code written on either side.
You need: The endpoint, the name of a tool that lists records and the name of a tool that fetches one, plus the argument names each expects.
Any SQL query
Whatever a read-only SELECT returns becomes memory, joins and filters included.
You need: A read-only connection string and two queries: one that lists, one that re-reads a single row.
Why there is no logo grid
An enterprise search platform sells connector count, and every connector is an auth flow, a rate limit, a schema and a permission model somebody maintains forever. Contextely takes the other side of that trade, and the cost is exactly what this page shows.
What Contextely does not reach
- A product with no MCP server and no queryable database. There is no scraping path.
- A record with no stable identifier. The re-fetch key is what makes freshness work.
- A file share. Contextely condenses records, not documents.
- A source's own permission model. Entitlement is expressed here, per source and per member.
10 systems are selectable in the app today. Connect one in the quickstart, or read how a source's credential is stored.