Setup

Connect AIeph

AIeph is an MCP server. Add it to your AI coding agent, and the agent checks the shared cache before your model generates an answer.

Endpoint: https://aieph.dev/mcp (Streamable HTTP)
Tool exposed: aieph_search
No account. No API key. Nothing to sign up for.

On a cache hit, your agent gets a known answer instantly. On a miss, AIeph stays silent and your model answers exactly as it normally would — it never guesses.

Config keys differ per client. Cursor and JetBrains use mcpServers; VS Code uses servers. Copy the block for your tool exactly.
Counting installs. The snippets below carry a random X-Aieph-Install ID generated in your browser just now. It exists so we can count how many people use AIeph — it is a random value, tied to nothing about you, and unrelated to the content of your queries. Delete that line and everything still works; you just won't be counted. See Privacy.
Cursor Claude Code VS Code JetBrains Verify Make it fire every time Uninstall Troubleshooting

Cursor

1. Create the config file.

2. Paste this in:

{
  "mcpServers": {
    "aieph": {
      "url": "https://aieph.dev/mcp"
    }
  }
}

3. Reload. Cursor picks up the server automatically. Open the MCP settings (Settings → MCP) and confirm aieph shows a green/connected status and that aieph_search appears in its tool list.

Claude Code

Option A — one command (recommended)

claude mcp add --transport http aieph https://aieph.dev/mcp

By default this registers AIeph for the current project only. To make it available in every project instead:

claude mcp add --transport http --scope user aieph https://aieph.dev/mcp

To share it with your team via a committed .mcp.json file:

claude mcp add --transport http --scope project aieph https://aieph.dev/mcp

Option B — edit .mcp.json by hand

Project root:

{
  "mcpServers": {
    "aieph": {
      "type": "http",
      "url": "https://aieph.dev/mcp"
    }
  }
}

Verify: run claude mcp list. aieph should show as connected. No sign-in step is needed (AIeph requires no authentication).

VS Code

Requires the GitHub Copilot extension. MCP tools only run in Agent mode.

1. Create the config file.

2. Paste this in (note the top-level key is servers, not mcpServers):

{
  "servers": {
    "aieph": {
      "type": "http",
      "url": "https://aieph.dev/mcp"
    }
  }
}

3. Reload the window. Command Palette → Developer: Reload Window.

4. Switch Copilot Chat to Agent mode. Open the Copilot Chat panel and change the mode dropdown from Ask to Agent — MCP tools do nothing in Ask mode. The aieph_search tool should now appear in the tools list.

JetBrains IDEs

IntelliJ IDEA, WebStorm, PyCharm, GoLand, RubyMine, and more. Requires the bundled AI Assistant plugin and a JetBrains AI subscription (a free trial works). Steps below are for AI Assistant 2026.x.

1. Open the MCP settings. Go to Settings (Ctrl + Alt + S) → Tools → AI Assistant → Model Context Protocol (MCP). (Or, in the AI chat, type / and choose Add Command.)

2. Add a server. Click the Add (+) button. The New MCP Server dialog opens.

3. Choose the HTTP connection type and paste this JSON:

{
  "mcpServers": {
    "aieph": {
      "url": "https://aieph.dev/mcp"
    }
  }
}

4. Set the server level — choose Global (all projects) or Project (this project only).

5. Click OK, then Apply. Applying starts the connection. Watch the Status column turn to connected. Click the icon in the Status column to see the available tools; aieph_search should be listed.

Older AI Assistant without native HTTP? If the dialog offers no HTTP option, use the stdio bridge instead: set the command to npx with args ["mcp-remote", "https://aieph.dev/mcp"]. Upgrading AI Assistant is the cleaner fix.

Verify it's working

In any of the four tools, ask your agent an error, library, or version question, for example:

How do I set up CORS middleware in Hono?

Your agent should call aieph_search before answering. If the cache has a match, you get it instantly. If not, the tool returns nothing and your model answers as usual — that's expected. A miss is silent by design; a wrong cached answer would be worse than none.

Make it fire every time

Adding the MCP only makes aieph_search available — whether the agent actually calls it before answering is left to the model. In short chats it usually does. In long or autonomous sessions the model drifts: as context grows, it quietly stops reaching for the cache. A one-line rule in your project instructions helps, but it's the first thing a crowded context forgets.

If your agent is Claude Code, you can make the reminder unconditional. A UserPromptSubmit hook re-injects the instruction on every turn, so it never falls out of context. Save this as hooks/aieph-reminder.mjs in your project:

let input = ''
process.stdin.on('data', (c) => { input += c })
process.stdin.on('end', () => {
  process.stdout.write(
    'Before implementing or answering a technical question, call the ' +
    '`aieph_search` MCP tool once first. On `found:false` proceed normally. ' +
    'Do NOT include user code, file contents, or chat history in the query.\n'
  )
  process.exit(0)
})

Then register it in .claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      { "hooks": [
        { "type": "command",
          "command": "node ${CLAUDE_PROJECT_DIR}/hooks/aieph-reminder.mjs" }
      ] }
    ]
  }
}

The hook's stdout is appended to the agent's context each turn, so the "check the cache first" instruction is present on every prompt instead of once at the top of a long session. It never blocks, never sees your code, and adds no latency worth measuring. Other clients don't expose an equivalent prompt hook yet — there the project-instruction rule is the best available lever.

Uninstall

AIeph stores nothing on your machine beyond the config lines you added — no daemon, no login, no local database. Removing it is just deleting those lines and restarting the client. Once removed, your agent simply answers the way it did before.

Claude Code

claude mcp remove aieph

If you registered it with a scope, remove it from that same scope:

claude mcp remove --scope user aieph
claude mcp remove --scope project aieph

Then confirm with claude mcp list — aieph should no longer be listed. If you added it by hand instead, delete the "aieph" entry from .mcp.json (or ~/.claude.json for user scope).

Cursor

Delete the "aieph" entry from .cursor/mcp.json (project) or ~/.cursor/mcp.json (global). If aieph was the only server, the file can be deleted outright. Reload Cursor and check Settings → MCP.

VS Code

Delete the "aieph" entry from .vscode/mcp.json, or from the user config (Command Palette → MCP: Open User Configuration). Then Command Palette → Developer: Reload Window.

JetBrains IDEs

Settings → Tools → AI Assistant → Model Context Protocol (MCP), select aieph in the list, click Remove (−), then Apply.

If you added the reminder hook

Remove the UserPromptSubmit entry you added to .claude/settings.json and delete hooks/aieph-reminder.mjs. Leaving the hook in place after removing the MCP would keep telling your agent to call a tool that no longer exists.

Nothing else to clean up. AIeph never wrote to your project, and queries are PII-masked and logged anonymously — see Privacy. Coming back later is the same one command as before.

Troubleshooting