A courier for AI agents.
Ship context directly between agents: from your session to a teammate's, or from Perplexity to Claude Code. No pasting, no share settings.
claude mcp add --transport http contexthog \ https://api.contexthog.com/mcp \ --header "Authorization: Bearer chog_sk_…"
// ~/.cursor/mcp.json
{
"mcpServers": {
"contexthog": {
"url": "https://api.contexthog.com/mcp",
"headers": { "Authorization": "Bearer chog_sk_…" }
}
}
}export CONTEXTHOG_API_KEY=chog_sk_… codex mcp add contexthog \ --url https://api.contexthog.com/mcp \ --bearer-token-env-var CONTEXTHOG_API_KEY
// .vscode/mcp.json
{
"servers": {
"contexthog": {
"type": "http",
"url": "https://api.contexthog.com/mcp",
"headers": { "Authorization": "Bearer chog_sk_…" }
}
}
}CONTEXTHOG_API_KEY=chog_sk_… npx -y @contexthog/mcp
Replace chog_sk_… with your key, then ask your agent to send something.
No key yet? ContextHog is invite-only for now. Anyone already on it can send you one.
Decision: token bucket in Redis, 100 requests a minute per API key.
Rejected: fixed windows. They let a client burst 200 requests across the minute boundary.
Open: per-org limits. Notes and sources below.
Send this to @omar?
Context goes where the work goes.
Send it to a teammate
Your agent packs up the session and sends it to their @handle. You skip "Claude says:" and post one line instead:
I sent you the context: contexthog.com/h/7F3K2AQXTheir agent opens the whole thing.
Share docs without share settings
Plans, specs, and notes travel as the message itself, up to 100,000 characters of Markdown. No access requests, and it works the same in Codex as in Claude.
Move a plan between your own agents
Draft it in Perplexity, build it in Claude Code. Your agents send to each other, so a plan doesn't have to be committed to the repo just to get across.
It speaks when you ask, and stays quiet otherwise.
Only on request
Agents are told to use ContextHog when you ask. It doesn't start a session by checking mail, report on its own progress, or offer to send things.
You approve every send
Clients that support MCP confirmation show you the recipient and message first. If you decline, the agent drops it and doesn't ask again.
First contact is screened
A message from someone new waits for your approval. Senders you haven't approved can't wake your agents.
Links carry no content
The link shows the sender, recipient, and time. Reading the message takes the recipient's own key.
Give your agents an address.
One command in the client you already use. Then ask your agent to send something.
claude mcp add --transport http contexthog \ https://api.contexthog.com/mcp \ --header "Authorization: Bearer chog_sk_…"
// ~/.cursor/mcp.json
{
"mcpServers": {
"contexthog": {
"url": "https://api.contexthog.com/mcp",
"headers": { "Authorization": "Bearer chog_sk_…" }
}
}
}export CONTEXTHOG_API_KEY=chog_sk_… codex mcp add contexthog \ --url https://api.contexthog.com/mcp \ --bearer-token-env-var CONTEXTHOG_API_KEY
// .vscode/mcp.json
{
"servers": {
"contexthog": {
"type": "http",
"url": "https://api.contexthog.com/mcp",
"headers": { "Authorization": "Bearer chog_sk_…" }
}
}
}CONTEXTHOG_API_KEY=chog_sk_… npx -y @contexthog/mcp
Replace chog_sk_… with your key, then ask your agent to send something.
No key yet? ContextHog is invite-only for now. Anyone already on it can send you one.
Context, delivered.