Hosted context engine for coding agents
The context your agent needs — routed, not dumped.
Your agent makes one call — get_context — and gets back only what's relevant
to the task: routed across your projects, precedence-resolved, provenance-stamped, and
trimmed to a token budget. Markdown is the source of truth. Hosted for your team — sign in and connect your agent in minutes.
claude mcp add --transport http contexel https://contexel.ai/mcp
Sign in to the console for your exact connect command (with your token). Works with Claude Code, Cursor, and any MCP client.
The problem
Bigger context windows didn't fix context.
Agents drown in stale CLAUDE.md files, context that bleeds across projects, and
conflicting instructions with no way to know which one wins. So you paste the same
background by hand, every session. Contexel makes context a query — not a copy-paste ritual.
How it works
One tool call. The right context back.
Sign in and connect your agent.
Open the console, then point Claude Code, Cursor, or any MCP client at your workspace endpoint — once.
You ask in normal words.
"What do we know about rate limiting?" The tools carry plain-English titles and the server briefs your agent the moment it connects — nothing to paste, no prompt gymnastics. A built-in help tool answers "what is this and what do I say?"
It gets only what matters.
Routed to the right realm, precedence-resolved, provenance-stamped, and packed to fit your token budget.
What it does
Context you can trust, by construction.
Realms
Work, personal, per-client — isolated by design. Context from one realm never bleeds into another. Separate stores, not a WHERE clause.
Precedence
When two instructions conflict, Contexel knows which wins — org, then team, then project — and shows you why.
Provenance
Every card carries where it came from. No anonymous context, and nothing is dropped silently.
Yours, and portable
Markdown is the source of truth; the index is derived and disposable, so you're never locked in. Every workspace is isolated, and you can turn on encryption at rest.
Deterministic
Rebuilding over the same files produces identical content digests — reproducible retrieval, no drift, no surprises.
Token-budgeted
You set the budget; Contexel packs the most relevant context to fit and flags anything it had to drop.
Simple on purpose
Every tool has a plain-English title, a built-in help tool answers "what is this and what do I say?", and your agent is briefed automatically at connect time. If you can describe your task, you can use it.
It learns — you approve
observe saves a quick note; propose drafts a lasting change a human accepts via review. Auto-commit is off by default, every commit carries provenance, and secrets are caught before anything is saved.
Bring what you have
Start from the docs you already have — a flat CLAUDE.md maps cleanly into routed context. No rewrite, no new format to learn; markdown stays the source of truth.
Built for teams
One shared context service for everyone.
Sign in and go.
Sign in with your browser — Google, GitHub, Microsoft, or SSO. Each workspace is isolated from every other, and you can turn on encryption at rest. No install, no setup.
Open the consoleConnect in one line.
Point Claude Code, Cursor, or any MCP client at your workspace endpoint. Agents connect with a minted token — or a browser sign-in (OAuth 2.1) where it's enabled — then call get_context, observe, and propose.
View on GitHubStraight talk
We tell you exactly what this is.
- The hosted service is custodial today. We can technically access your data — it's stored under platform-level controls, and when you turn on workspace encryption we keep a copy of the key to run indexing and search. Bring-your-own-key is on the roadmap — we'll say so when it ships, not before.
- No proprietary format. Your content is plain markdown and the index is derived and disposable — nothing proprietary to migrate, nothing to reverse-engineer.
- Plain claims only. If we can't back it, we don't say it — no inflated crypto language, no marketing sleight of hand.
Get started
From sign-in to routed context in minutes.
Open the console and sign in with your browser — Google, GitHub, Microsoft, or SSO. No install; your workspace is ready immediately.
Open the consoleclaude mcp add --transport http contexel \ https://contexel.ai/mcp \ --header "Authorization: Bearer <YOUR_TOKEN>"
Mint a token in the “Connect your agent” panel and it shows this line filled in — for Claude Code, Cursor, or any MCP client. Where browser sign-in (OAuth 2.1) is enabled, add the endpoint with no token and sign in from your editor.
Then just talk to it: ask "what do we know about rate limiting?" and your agent calls get_context; say "remember this for next time" and it calls observe — a human approves anything lasting via review. Not sure what to say? There's a help tool for that.
Give your agent the right context.
Hosted for your team — sign in and connect in minutes.