ailtir-mcp

An MCP server that gives Claude direct access to Ailtir's Knowledge Base platform. Upload documents, build a Bedrock-powered knowledge base, and run RAG chat — all from within your AI assistant.

Prerequisites

• Python 3.13+ (or uv — no separate Python install needed with uvx)
• An Ailtir account
• An AILTIR_MCP_API_TOKEN for authenticated knowledge-base and profile tools. The plugin reporting tools do not require a token.

Installation

uvx runs the server in an isolated environment with no install step — it is the recommended approach. You need uv installed (curl -LsSf https://astral.sh/uv/install.sh | sh).

Claude Code

claude mcp add ailtir uvx ailtir-mcp \
  -e AILTIR_MCP_API_TOKEN=your-secret-here

Or add manually to .claude/settings.json:

{
  "mcpServers": {
    "ailtir": {
      "command": "uvx",
      "args": ["ailtir-mcp"],
      "env": {
        "AILTIR_MCP_API_TOKEN": "your-secret-here"
      }
    }
  }
}

Claude Desktop

Add the following to your claude_desktop_config.json (~/Library/Application Support/Claude/ on macOS, %APPDATA%\Claude\ on Windows):

{
  "mcpServers": {
    "ailtir": {
      "command": "uvx",
      "args": ["ailtir-mcp"],
      "env": {
        "AILTIR_MCP_API_TOKEN": "your-secret-here"
      }
    }
  }
}

Replace your-secret-here with the secret you copied in Step 1, then restart Claude Desktop.

Persistent install (optional)

If you prefer ailtir-mcp installed as a persistent global tool:

uv tool install ailtir-mcp

Then use ailtir-mcp as the command instead of uvx ailtir-mcp. pip install is not recommended — it installs into the active Python environment and can cause dependency conflicts.

Tools

Once connected, your AI assistant has access to authenticated Ailtir platform tools plus two public plugin reporting tools.

plugin_report_usage (public)

Reports an anonymous Ailtir plugin skill invocation. It does not require AILTIR_MCP_API_TOKEN and does not collect a stable installation or workspace identifier.

plugin_report_usage(skill_name: string, plugin_version: string) → PluginReportResult

plugin_feedback (public)

Submits an anonymous 1–10 rating, optional short reason, workflow name, and up to three structured follow-up answers. It does not require AILTIR_MCP_API_TOKEN.

plugin_feedback(rating: int, plugin_version: string, ...) → PluginReportResult

Both tools return status: submitted|failed; reporting failures are visible but do not raise an MCP tool error.

upload

Uploads a ZIP archive of documents to Ailtir storage and returns a kb_id.

upload(file_path: string) → kb_id: string

Returns a kb_id that you pass to analyse, list, and chat.

analyse

Unzips the uploaded archive and builds an AWS Bedrock Knowledge Base from its contents. Triggers the full ingestion pipeline — this typically takes a few minutes.

analyse(kb_id: string) → status: string

list

Lists all knowledge bases associated with your Ailtir account.

list() → string

Returns a formatted list of knowledge bases, each showing name, kb_id, and status (e.g. ready, analysing, failed).

chat

Asks a question answered using documents in a given knowledge base (retrieval-augmented generation via AWS Bedrock).

chat(kb_id: string, question: string) → answer: string

Getting your secret

1. Sign in to the Ailtir app
2. Click your avatar or name to open the Account page
3. Find the Secrets cell and click Reveal
4. Copy the MCP Key — it starts with amcp_
5. Paste it as AILTIR_MCP_API_TOKEN in your client config (see Installation)

Links

• Ailtir
• MCP specification
• uv
• Claude Desktop
• api-mcp service (the backend this server talks to)