Self-hosted AI workspace: semantic RAG document search, long-term memory, and AI chat via MCP.
Scored across 10 tools
Most tools target clearly distinct resources and actions (chat listing, memory add/search, file ingestion, RAG search, full chat pipeline). The only real overlap is rag_search vs rag_similar, which both return document chunks but differ by query type (semantic query vs message_id), and descriptions mostly resolve this.
There is a recognizable resource-action pattern in memory_add, memory_search, rag_search, rag_similar, and synaplan_chat, but file_ingest flips to noun-verb while the retrieval tools use verb-first (get_messages, list_chats). It remains readable, but conventions are mixed.
Ten tools is well within the ideal range and each covers a distinct capability: listing resources, reading messages, ingesting documents, RAG retrieval, memory management, and the main chat pipeline. No obvious filler tools.
Core workflows (ingest, search, chat, memory add/search, list chats/messages/assistants/prompts) are covered, but lifecycle operations are notably missing: there is no way to list or delete ingested documents, and no update/delete for long-term memories. Agents can work around some gaps but will hit dead ends for document and memory management.
Add a text document to the authenticated user's knowledge base. The content is chunked, embedded, and stored so it becomes retrievable via rag_search.
Inputs: name (required), content (required), group_key.
Return the messages of one of the user's chats, in chronological order.
Inputs: chat_id (required), limit.
List assistants the authenticated user may use from a connected app. Each row includes slug, name, description, version and origin.
Inputs: no named parameters.
List the authenticated user's conversations (most recently updated first).
Inputs: limit.
List the user's available Synaplan task prompts (internal tools:* prompts excluded).
Inputs: no named parameters.
Store a new long-term memory for the authenticated user — a durable fact or preference that Synaplan will recall in future conversations.
Inputs: key (required), value (required), category (required).
Search the authenticated user's stored long-term memories (preferences, facts, context) using semantic similarity.
Inputs: query (required), limit, category.
Semantic search across the authenticated user's vectorized documents and knowledge base. Returns the most relevant text chunks with similarity scores.
Inputs: query (required), limit, group_key, min_score.
Find document chunks similar to a given source message/document id (the message_id returned by rag_search).
Inputs: message_id (required), limit.
Send a message through Synaplan's full AI pipeline — intent classification, web search, RAG over the user's documents, long-term memories, and inference — and return the synthesized answer. This is richer than a raw model call: it uses the authenticated account's knowledge base and memories.
Inputs: message (required), agentId, chat_id.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Private persistent memory for Claude, ChatGPT & Gemini via MCP - semantic search, zero-code setup.
One memory, every AI. A shared, user-owned markdown memory your AI clients read and write over MCP.
Persistent memory, knowledge graph, task management and more for AI agents via MCP.
Hosted MCP memory: save sessions/decisions once, search from Claude, Cursor, ChatGPT. EU-hosted FTS.
Add one secure layer between your agents and this server.
