anansi-mcp

MCP server for Anansi — persistent, synthesized memory for any MCP client: Claude Desktop, Claude Code, Cursor, Windsurf, and anything else that speaks the Model Context Protocol.

Unlike raw vector-search memory, Anansi runs a synthesis pass over everything you store and returns a curated profile — stable facts, current context, and relevant memories — ready to use directly.

Quick start

npx -y anansi-mcp

That's it — no install step. The server reads its configuration from two environment variables:

| Variable | Required | Description | |---|---|---| | ANANSI_API_KEY | yes | Your Anansi API key (ans_...) — create one in the developer portal | | ANANSI_BASE_URL | no | Override the API endpoint (defaults to the hosted Anansi API) |

Setup

Claude Desktop

Add to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "anansi": {
      "command": "npx",
      "args": ["-y", "anansi-mcp"],
      "env": {
        "ANANSI_API_KEY": "ans_your_key_here"
      }
    }
  }
}

Claude Code

claude mcp add anansi --env ANANSI_API_KEY=ans_your_key_here -- npx -y anansi-mcp

Cursor

Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global):

{
  "mcpServers": {
    "anansi": {
      "command": "npx",
      "args": ["-y", "anansi-mcp"],
      "env": {
        "ANANSI_API_KEY": "ans_your_key_here"
      }
    }
  }
}

Windsurf

Add the same block to ~/.codeium/windsurf/mcp_config.json.

Tools

remember

Save a memory. Content is chunked, embedded, and synthesized into the user's profile asynchronously.

| Argument | Type | Required | Description | |---|---|---|---| | content | string | yes | The memory to store — a fact, preference, note, or conversation excerpt | | userId | string | yes | Stable identifier for the user this memory belongs to | | sourceType | conversation | document | note | meeting | custom | no | What kind of content this is (default: conversation) |

recall

Retrieve the synthesized profile for a user: stable facts, current context, and the memories most relevant to the query.

| Argument | Type | Required | Description | |---|---|---|---| | query | string | yes | What you want to know — used for relevance search | | userId | string | yes | Stable identifier for the user whose memory to retrieve |

Example output:

## Stable facts
- Prefers TypeScript over Python
- Senior engineer at a fintech startup

## Current context
- Debugging a webhook system this week

## Relevant memories
- Mentioned the webhook retries fail silently when Redis is saturated

Development

pnpm install
pnpm --filter anansi-mcp build
ANANSI_API_KEY=ans_... node packages/mcp/dist/index.js

The server speaks MCP over stdio. Test it interactively with the MCP Inspector:

npx -y @modelcontextprotocol/inspector npx -y anansi-mcp

License

MIT