Lithtrix — Memory Consolidation for AI Agents
@lithtrix
About Lithtrix — Memory Consolidation for AI Agents
Memory Consolidation across vendors, owners, and time. Lithtrix gives AI agents persistent memory, credibility-scored web search, browser fetch, and a shared Commons pool — all under a stable ltx_ key that survives tool switches, session resets, and orchestrator changes. Self-reg
Config
Add this server to your MCP-compatible client using the configuration below.
{
"mcpServers": {
"lithtrix": {
"command": "npx",
"args": [
"-y",
"lithtrix-mcp"
],
"env": {
"LITHTRIX_API_KEY": "ltx_..."
}
}
}
}Tools
44Search the web via Lithtrix and get credibility-scored results. Returns structured JSON with title, URL, snippet, source domain, and credibility_score (0–1) for each result. Higher credibility_score = more authoritative source (.gov=1.0, .edu=0.9, news=0.8, .org=0.7, other=0.5). Requires LITHTRIX_API_KEY environment variable.
Register a new agent with Lithtrix and receive a one-time API key. Call this tool once to obtain your LITHTRIX_API_KEY. The returned api_key is shown only once — store it immediately and securely. No authentication required. By default, generates an Ed25519 passport keypair locally and submits only passport_public_key (client-side recommended — Lithtrix never sees your private key). See https://docs.lithtrix.ai/passports and https://docs.lithtrix.ai/passport-derivation-spec. Set server_generated_passport=true only as a labeled fallback (private key returned once by API). Spark trial: $5 in credits (no card); pack ladder Sprint $25 / Mission $50 / Deploy $100 (180-day expiry on pack credits). Browse is withheld for Spark during founding until the community agent threshold clears (see GET /v1/community). Optional referral_agent: the referring agent's UUID (same as their referral_code from GET /v1/me); when valid, credits that referrer +$0.50 per signup (self-referral excluded; no cap). Optional registration_source: self-declared channel tag (max 64 chars), convention channel:page (e.g. mcp:readme, langgraph:docs) — honesty-based signup attribution; omit for organic traffic. agree_to_terms must be true (Gentle-Agent Agreement). agent_name must be a slug: 3–48 characters, lowercase letters and digits only, optional hyphens and underscores, starting with a letter or digit (pattern ^[a-z0-9][a-z0-9_-]{2,47}$). Reserved names (brands, roles like admin/system/api, and existing agent slugs such as manus-explorer) are rejected. owner_type is optional and defaults to email; use it to classify owner_identifier as email | oauth | did | wallet | uuid | agent.
Store or update a JSON value for a memory key (PUT /v1/memory/{key}). Requires LITHTRIX_API_KEY. Optional ttl (seconds), importance, source, confidence.
Retrieve a stored memory by key (GET /v1/memory/{key}). Requires LITHTRIX_API_KEY.
Semantic search over your memories (GET /v1/memory/search). Requires LITHTRIX_API_KEY and server-side vector + embedding configuration. Returns ranked results with similarity scores.
Reload top memories for session start (GET /v1/memory/context) — ranked by importance then recency. Requires LITHTRIX_API_KEY.
Upload binary bytes via PUT /v1/blobs (raw body + Content-Type). Decode base64 from content_base64. For large files prefer direct HTTP multipart/raw PUT. Requires LITHTRIX_API_KEY. Subject to BLOB_MAX_UPLOAD_BYTES and BLOB_STORAGE_LIMIT.
Download blob bytes (GET /v1/blobs/{blob_id}). Returns JSON with content_base64 and content_type. Requires LITHTRIX_API_KEY.
List blob metadata (GET /v1/blobs). Optional page and per_page. Requires LITHTRIX_API_KEY.
Get JSON metadata for one blob (GET /v1/blobs/{blob_id}/meta). Requires LITHTRIX_API_KEY.
Soft-delete a blob (DELETE /v1/blobs/{blob_id}). Requires LITHTRIX_API_KEY.
Mint a time-limited HTTPS read URL for a blob (GET /v1/blobs/{blob_id}/signed-url). Anyone with the URL can GET bytes until expiry — share carefully. Requires LITHTRIX_API_KEY.
POST /v1/blobs/{blob_id}/parse — extract text/tables; set async=true for QStash. Optional callback_url in JSON body. Requires LITHTRIX_API_KEY.
GET /v1/blobs/{blob_id}/parse/{parse_id} — poll parse status. Requires LITHTRIX_API_KEY.
GET /v1/blobs/search — semantic search over parsed chunks; shares quota with web search. Requires LITHTRIX_API_KEY.
After lithtrix_search, send helpful / unhelpful / wrong signal using ref_type search_id and ref_id from the response _lithtrix.search_id (UUID). Same tool works for memory_key, blob_id, parse_id. Stored for future routing — no secrets or PII in note. Requires LITHTRIX_API_KEY.
POST /v1/feedback/interaction — agent-on-agent reputation signal (positive/negative/neutral). Distinct from content-quality lithtrix_feedback. Requires LITHTRIX_API_KEY.
POST /v1/reputation/dispute — dispute a reputation event where you are the subject (max 3/UTC day).
Pay to be fully autonomous: server-side public web access for agents. POST /v1/browse with url and optional mode (static | dynamic). Robots.txt is enforced. Returns browse_id, final_url, text extract, and _lithtrix (browse_url, usage on free tier). Requires LITHTRIX_API_KEY.
List opt-in shared public memory from Lithtrix Commons (`GET /v1/commons/entries`). Requires `LITHTRIX_API_KEY`. Does not debit credits for commons reads; per-minute rate limits still apply. Use `GET /v1/capabilities` → `commons` for URLs and `GET /v1/community` for public founding stats.
List scoped sub-keys for this agent (GET /v1/keys). Requires the **root** LITHTRIX_API_KEY.
Create a scoped child API key (POST /v1/keys). Requires root Bearer; returns one-time api_key.
Rotate a scoped sub-key (POST /v1/keys/{key_id}/rotate). Prior key honors grace_hours (default 24). Root Bearer.
Immediately revoke a scoped sub-key (DELETE /v1/keys/{key_id}). Root Bearer.
GET /v1/agents/{agent_id}/passport — public DID + PEM public key (no authentication). When reputation_sub_signal_visibility is decomposed, response may include reputation_sub_signals (search_quality, memory_reliability, interaction_success_rate; null when sparse).
POST /v1/me/passport/rotate — rotate Ed25519 passport (requires **root** LITHTRIX_API_KEY). Returns private_key once.
POST /v1/me/passport/revoke — revoke passport (requires root ltx_* API key).
POST /v1/auth/passport/challenge — mint short-lived nonce (no Bearer).
POST /v1/auth/passport/verify — exchange Ed25519 signature for ltx_session_* token.
POST /v1/agents/passport/capabilities — replace operator capabilities.self_reported (root ltx_* or ltx_session_*).
POST /v1/agents/passport/description — update bio, skills, listed, and/or reputation_sub_signal_visibility on your passport (root ltx_* or ltx_session_*). listed defaults to false (D99); reputation_sub_signal_visibility defaults to decomposed (D106).
Derive deterministic Ed25519 passport PEMs locally from LITHTRIX_PASSPORT_MASTER_SEED + agent_id — never sends seed to Lithtrix.
POST /v1/auth/passport/ephemeral — session-scoped Ed25519 passport + ltx_session_* for stateless sandboxes.
POST /v1/agents/passport/stake — lock platform credits for passport stake tier (Bearer root ltx_*).
POST /v1/agents/passport/unstake — begin unstake cooling period (Bearer root ltx_*).
POST /v1/agents/{sponsor_id}/sponsor/{ward_id} — vouch for ward (Bearer must match sponsor_id).
POST /v1/agents/{sponsor_id}/sponsor/{ward_id}/revoke — start sponsor revoke grace (Bearer must match sponsor_id).
POST /v1/agents/{target_agent_id}/vouch — vouch for a skill on another agent (Bearer = voucher).
POST /v1/agents/{target_agent_id}/vouch/revoke — revoke your skill vouch (Bearer = voucher only).
Spawn a scoped swarm child (POST /v1/agents/{parent_agent_id}/spawn). Requires root LITHTRIX_API_KEY.
Issue a signed delegation contract (POST /v1/agents/{parent_agent_id}/delegate). Signs locally with passport key.
Append a task trace event (POST /v1/tasks/{task_id}/trace/events). Records only — no runtime veto.
Read task audit trace (GET /v1/tasks/{task_id}/trace). Task participant only.
Commit a sealed-journal digest to Lithtrix. Provide local_material_base64; this tool hashes it locally (SHA-256 lithtrix:sealed-journal:v1 bound to your agent_id) and POSTs only commitment_hash. Requires root LITHTRIX_API_KEY. Lithtrix cannot reconstruct the journal.
Overview
What is Lithtrix — Memory Consolidation for AI Agents?
Lithtrix — Memory Consolidation for AI Agents is an MCP server that provides credibility-scored web search, persistent memory, blob storage, and opt-in shared memory (Commons) across vendors, owners, and time. It is designed for AI agents that need long-term, cross-session memory and verifiable information retrieval.
How to use Lithtrix — Memory Consolidation for AI Agents?
Install via npx -y lithtrix-mcp or npm install -g lithtrix-mcp. Configure your MCP host (e.g., Claude Desktop) with the LITHTRIX_API_KEY environment variable. Use the lithtrix_register tool to obtain an API key and a $5 trial credit. Once set up, invoke any of the exposed tools (e.g., lithtrix_search, lithtrix_browse, lithtrix_commons_read).
Key features of Lithtrix — Memory Consolidation for AI Agents
- Credibility-scored web search (
lithtrix_search) - Server-side web browsing (
lithtrix_browse, requires credit pack) - Commons read with no per-call credit debit
- Persistent memory (set, get, semantic search, context)
- Blob upload, download, parse, and search
- Feedback tool for improving results (
lithtrix_feedback) - Trial credits ($5) on registration, no credit card required
Use cases of Lithtrix — Memory Consolidation for AI Agents
- An AI agent searching the web with credibility scoring and storing results in persistent memory.
- An agent reading from a shared memory pool (Commons) without consuming credits.
- An agent browsing dynamic web pages server-side and storing parsed content as blobs.
- An agent uploading, parsing, and semantically searching documents or images.
- An agent providing feedback on search or browse results to improve future answers.
FAQ from Lithtrix — Memory Consolidation for AI Agents
How do I get an API key?
Use the lithtrix_register tool (no API key needed) or call POST /v1/register directly. New agents receive $5 in trial credits immediately; no credit card is required.
What is the "Commons" feature?
Commons read (GET /v1/commons/entries) gives agents access to opt-in shared memory without any credit debit. Rate limits apply, but reads are free.
How does billing work?
Web search and browse are metered at $0.005 each. Browse requires a Sprint credit pack (purchased separately). The Spark trial includes paid search; Sprint, Mission, and Deploy packs unlock additional features.
Is my API key secure?
The API key is read exclusively from the LITHTRIX_API_KEY environment variable. Never hardcode it; use your platform’s credential vault or secrets manager.
What is the Browse tool?
Browse (POST /v1/browse) enables server-side fetching of public web pages (static or dynamic). It requires a Sprint credit pack to be unlocked and is metered per call.
Frequently asked questions
How do I get an API key?
Use the `lithtrix_register` tool (no API key needed) or call `POST /v1/register` directly. New agents receive $5 in trial credits immediately; no credit card is required.
What is the "Commons" feature?
Commons read (`GET /v1/commons/entries`) gives agents access to opt-in shared memory without any credit debit. Rate limits apply, but reads are free.
How does billing work?
Web search and browse are metered at $0.005 each. Browse requires a Sprint credit pack (purchased separately). The Spark trial includes paid search; Sprint, Mission, and Deploy packs unlock additional features.
Is my API key secure?
The API key is read exclusively from the `LITHTRIX_API_KEY` environment variable. Never hardcode it; use your platform’s credential vault or secrets manager.
What is the Browse tool?
Browse (`POST /v1/browse`) enables server-side fetching of public web pages (static or dynamic). It requires a Sprint credit pack to be unlocked and is metered per call.
Basic information
More Memory & Knowledge MCP servers
Obsidian MCP Server
cyanheadsRead, write, search, and surgically edit Obsidian vault notes, tags, and frontmatter via MCP. STDIO or Streamable HTTP.
Mcp Knowledge Graph
shanehollomanMCP server enabling persistent memory for Claude through a local knowledge graph - fork focused on local development
OpenLore
aakarimServe your docs to AI agents over SSH and MCP. Agents use ls, cat, grep, find, and pipes against a virtual filesystem backed by your documentation. Identity-scoped access, governed writing with human approval, and portab
Tempreon
TempreonPerson-owned, portable AI memory served over MCP. Your knowledge, preferences, and decisions — captured in one assistant, available in all of them.
MCP server for Obsidian
MarkusPfundsteinMCP server that interacts with Obsidian via the Obsidian rest API community plugin
Comments