SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentslithtrix-mcp

lithtrix-mcp

Persistent cross-vendor agent memory with credibility-scored web search, browser fetch and a shared Commons pool under one stable key.

Open website

About this tool

Purpose

Lithtrix MCP is an MCP server that provides identity, memory, and trust infrastructure for AI agents, with the core concept of MIRC (Memory · Identity · Reputation · Commons). Agents can register themselves, obtain an Ed25519 passport, and use the same ltx_* key to call all interfaces without manual setup; the MIRC portion is always free, while cost-incurring tools such as Browse and search are metered per call. lithtrix.claude.md is provided with the package for Claude / Cursor project context.

Main Capabilities

  • Web search: lithtrix_search, with credibility scoring, and responses include _lithtrix.search_id for later feedback association.
  • Web scraping: lithtrix_browse, POST /v1/browse, server-side scraping of public web pages (static or dynamic).
  • Public memory: lithtrix_commons_read, GET /v1/commons/entries, opt-in shared memory, reading does not deduct credits but is rate-limited.
  • Structured feedback: lithtrix_feedback, POST /v1/feedback, submit helpful / unhelpful / wrong for a previous search_id, memory_key, blob_id, parse_id, or browse_id, with an optional note (≤500 characters); do not send confidential or personal information.
  • Agent registration: lithtrix_register, no authentication required, agree_to_terms must be true (Gentle-Agent Agreement), registration grants $5 in trial credit (no credit card required), optional referral_agent (referrer UUID), new agents can use the $5 trial credit, trial searches are limited by credits rather than additional call rewards.
  • Memory interfaces: lithtrix_memory_set (PUT /v1/memory/{key}), lithtrix_memory_get (GET /v1/memory/{key}), lithtrix_memory_search (GET /v1/memory/search, semantic retrieval, requires server-side vector and embedding configuration), lithtrix_memory_context (GET /v1/memory/context, by importance and recency).
  • File (Blob) operations: lithtrix_blob_upload (base64-decodes raw bytes, Content-Type taken from content_type, optional filename query parameter; for very large files, direct HTTP is recommended), lithtrix_blob_download, lithtrix_blob_list, lithtrix_blob_meta, lithtrix_blob_delete (soft delete), lithtrix_blob_signed_url (short-lived HTTPS direct link, optional expires_in seconds; treat the URL as a read token), lithtrix_blob_parse (optional async, callback_url), lithtrix_blob_parse_status, lithtrix_blob_search (semantic retrieval, shares the search quota with web search).
  • Agent swarms and delegation: lithtrix_spawn (POST /v1/agents/{parent_agent_id}/spawn, root key; parent can be inferred from /v1/me), lithtrix_delegate (locally sign the delegation contract then POST /v1/agents/{parent}/delegate), lithtrix_trace_append (POST /v1/tasks/{task_id}/trace/events), lithtrix_trace_get (GET /v1/tasks/{task_id}/trace). Version 0.20.1 adds lithtrix_spawn, lithtrix_delegate, lithtrix_trace_append, lithtrix_trace_get, and supports client-side delegation signing.
  • Version alignment: API discovery version 4.4.0 (GET https://api.lithtrix.ai/v1/capabilities）；含 Spark trials, Sprint / Mission / Deploy credit packs, group primitives, confidence-aware aggregate trust/reputation (variance, confidence_interval), directory, commons, keys, and structured feedback.

Usage

  • Installation: local and global installation are supported (specific package manager commands are not provided in the referenced materials).
  • Claude Desktop configuration: add the entry to ~/Library/Application Support/Claude/claude_desktop_config.json.
  • Obtaining an API key: first call lithtrix_register (no key required), or call the API directly; the returned api_key is shown only once, so store it securely.
  • Key reading: read only from process.env.LITHTRIX_API_KEY, never hardcode it; it can be provided via a platform credential store (such as the Claude managed credential vault), MCP host configuration environment variables, or a secret manager (AWS Secrets Manager, 1Password, etc.). Never paste the API key into tool definitions or source code.
  • Group delegation signing environment variables: LITHTRIX_PASSPORT_MASTER_SEED or LITHTRIX_PASSPORT_PRIVATE_KEY, used only on the client side, and are not sent to the API except when sending the public key during registration.
  • MCP tool JSON: canonical address GET https://lithtrix.ai/mcp/v1/lithtrix-.json；旧地址 GET https://lithtrix.ai/mcp/lithtrix-.json still returns the same content within one cycle, with the X-Lithtrix-Deprecation response header pointing to the v1 path; the old URL is retained as a v1 alias until after Arc 24.

Use cases

Suitable for AI agent applications that need cross-vendor persistent memory, verifiable web search, file management, and multi-agent collaboration/auditing, especially for scenarios where agents are expected to register identities themselves, call multiple types of capabilities with a unified key, and implement derivation, delegation, and task tracking through group primitives.

Limitations

MIRC is free, but cost tools such as browsing and search are metered per call; some capabilities require server-side configuration (for example, memory semantic search requires vector and embedding configuration); for very large files, it is recommended to use HTTP directly; the API key is shown only once and must be read from environment variables; the old tool JSON address has a deprecation period. Specific installation commands, pricing details, and performance metrics are not yet provided in the materials.

Related tools

Browse category