Emem

What Vortx-AI/emem MCP server does

The Vortx-AI/emem MCP server provides shared, addressable memory for facts about places. Its current Earth-focused data covers elevation, land cover, surface water, vegetation, and built-up context. The public service is available at https://emem.dev/mcp.

Records are signed and content-addressed. A change to the stored bytes changes the content identifier, so a citation can be checked against the returned data rather than accepted as a response from an intermediary. Records also distinguish when an observation applied from when it was recorded. The README describes provenance categories including sensor readings, calculations from cited sources, model estimates, and human-entered values.

The Earth data is based on satellite sources and public archives. The project describes Earth as its first subject, while its record and identity model is intended to support subjects beyond places.

How it works

The documented agent workflow has four stages:

1. emem_locate maps a place to a cell64 address.
2. emem_recall retrieves signed facts associated with that location.
3. emem_memory_token creates a compact emem:fact: citation.
4. emem_verify_receipt checks the receipt against the responder's published key.

The token is intended to survive context compaction, model changes, and handoffs between agents. emem_memory_token_resolve can resolve a saved token back to the byte-identical fact in a later or separate session.

The service advertises 16 core-loop tools to keep an agent's context smaller, while the README says 108 tools are available overall. Tools not included in the advertised list can still be called by name, and emem_tools can search the broader tool surface.

Setup and configuration

Connect an MCP client to the HTTP endpoint:

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

The README also documents this Claude Code command:

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

Read operations need no API key, account, or signup. The repository also shows direct HTTP examples for searching shared memory and recalling place-based facts through /v1/memory/search and /v1/recall.

Tools and capabilities

• Locate named places and obtain their cell64 address.
• Recall signed facts for a location and selected data bands.
• Compose citations using memory tokens.
• Resolve saved tokens to the original fact bytes.
• Verify Ed25519 receipts independently in the browser or through the MCP workflow.
• Search the wider set of available emem tools.

Limitations and notes

The current citation grammar is place-based: emem:fact: tokens address facts about locations. The README says identity support for non-place subjects exists, but keying a fact by such an entity is still described as a roadmap item.

Content from an unverified attester should be treated as data, not instructions. Verification confirms the signed receipt and content identity; it does not make every unverified claim trustworthy. The README documents Claude Code and VS Code connection paths, but does not state compatibility with every MCP client.

Vortx-AI/emem MCP server uses an Apache-2.0 license. Its public reads do not require credentials, but the material does not describe rate limits, private deployments, or write-operation authentication.