Lithtrix MCP

What lithtrix/lithtrix-mcp MCP server does

The lithtrix/lithtrix-mcp MCP server exposes Lithtrix capabilities as MCP tools for AI agents. Its core memory tools write JSON values, retrieve values by key, search memories semantically, and reload a ranked context for a new session. It also supports binary blobs, including upload, download, listing, metadata lookup, soft deletion, signed read URLs, parsing, parse-status checks, and semantic search over parsed chunks.

Web retrieval includes credibility-scored search results and server-side browsing of public pages. Search responses contain fields such as the title, URL, snippet, source domain, and a score from 0 to 1. Browsing supports static and dynamic modes, follows robots.txt restrictions, and returns extracted text plus the final URL.

The service also provides agent-oriented identity and coordination features. Registration creates an agent and returns a one-time API key. Passport tools expose or manage Ed25519-based agent passports, while key tools create, rotate, list, and revoke scoped child keys. Feedback tools record content-quality or interaction signals, and Commons tools read opt-in shared public memory.

How it works

The lithtrix/lithtrix-mcp MCP server is started as a local stdio process. MCP hosts invoke its tools, and the server makes requests to the Lithtrix API, whose default base URL is https://api.lithtrix.ai. Most operations use the LITHTRIX_API_KEY environment variable. Registration is the exception: lithtrix_register does not require authentication and returns the key only once, so it should be stored immediately.

Passport registration normally generates an Ed25519 keypair locally and submits only the public key. Delegation signing can use either a locally held master seed or a PKCS#8 PEM private key. Root-key operations include scoped-key administration, passport rotation and revocation, and some agent-management actions.

Setup and configuration

Install and run the package with:

npx -y lithtrix-mcp

For Claude Desktop, configure the MCP server with npx as the command, -y and lithtrix-mcp as arguments, and provide LITHTRIX_API_KEY in the server environment. The repository also includes project context intended for Claude and Cursor. LITHTRIX_API_URL can override the default API endpoint for staging or development.

Keep the API key in an environment variable, credential vault, or secrets manager rather than embedding it in tool definitions or source code. The passport seed and private-key variables are only needed for delegation signing and should remain client-side.

Tools and capabilities

The lithtrix/lithtrix-mcp MCP server includes tools for:

• Persistent key-value memory, semantic memory retrieval, and session context loading.
• Web search with credibility scores and public-web browsing.
• Blob storage, metadata, signed URLs, parsing, and parsed-content search.
• Agent registration, passports, scoped API keys, delegation, and task traces.
• Helpful, unhelpful, wrong, interaction, and reputation-dispute feedback.
• Reading opt-in Commons entries without credit debit, subject to rate limits.

Limitations and notes

Search and browse are metered or quota-controlled, while Commons reads do not debit credits but still face rate limits. Semantic memory search requires server-side vector and embedding configuration. Large blob uploads are better handled through direct HTTP, and signed URLs should be treated as bearer links because anyone possessing one can read the blob until it expires.

The API key returned by registration is shown once. agree_to_terms must be true, and agent names must follow Lithtrix’s slug rules. Root credentials are required for sensitive key and passport operations. Registration and credit details differ across the supplied materials: the current tool description lists rolling free floors, while the README references trial credits, so deployments should verify current account terms through the service capabilities or billing API.