MCP SERVER · OFFICIAL PROJECTPersistent memory for an agent — store with importance weighting, recall semantically or by BM25, group a run into a session, and expand a result along knowledge-graph edges.
LABSCO SUMMARY
The core profile is about 2,964 tokens of schemas; the all profile reaches 87 tools and roughly 16,212. Rather than making you choose once, discovery is a tool: search the catalogue for a name, then load that one schema. The deletion tools show similar care — batch_forget refuses to run without a filter, and the descriptions point you at lowering importance instead of deleting when what you want is suppression rather than removal. Note that this package is only the client; the memory server is a separate thing you have to be running.
The MCP front end for a Dakera memory server you run yourself. Fourteen tools load by default; the rest of the catalogue stays behind a discovery pair so a session does not begin by spending its context on schemas.
A Dakera memory server of your own — the MCP process is a client, not the store. Run it from the published Docker image, or use the deploy repository's compose file for persistent storage, then point DAKERA_API_URL and DAKERA_API_KEY at it. The client itself installs however you prefer: npx or a global npm install on Node.js 18+, Homebrew, cargo, a Docker image, or a pre-built binary for macOS, Linux or Windows. DAKERA_MCP_PROFILE decides how many tools appear in tools/list, and it can also be set per request.
One command plus a key — npm install -g @dakera-ai/dakera-mcp, then supply credentials
USE CASES
Store the decision with an importance weight, then recall it semantically next time the topic comes up.
Keyword search beats vector search when the thing you remember is an error code or a name.
Open a session, let everything store under it, and close it with a summary you can read back.
Build the graph from a seed memory, or expand a recall a few hops along its edges.
Delete by tag or time window rather than one ID at a time — but a filter is mandatory.
Extract the entities, topics and a summary first, then store those rather than the raw blob.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
dakera_batch_forgetBulk-delete memories matching filter criteria: tags, importance range, time window, or memory type. At least one filter is required to prevent accidental full-agent wipe. Deletion is permanent — use dakera_memory_importance to lower importance scores instead of deleting.
dakera_batch_recallFilter-based memory listing by tags, importance range, time window, type, or session. Prefer over dakera_recall when semantic search is not needed. At least one filter required.
dakera_discover_toolsSearch the Dakera tool catalog by keyword or tier (core/power/admin/meta) and return names and one-line summaries without loading full schemas. Call this first to find relevant tools, then use dakera_load_tools to fetch only the schemas you need — avoids loading the full catalog upfront.
dakera_extractExtract structured information (entities, topics, key phrases, summary) from arbitrary text using the configured provider hierarchy: per-request override → namespace default → server default → GLiNER local. Supported providers: gliner (zero-config local ONNX), openai, anthropic, openrouter, ollama, none.
dakera_forgetPermanently delete memories by ID or tag. Provide memory_ids for exact removal or tags to bulk-delete all memories sharing those tags. Deletion is immediate and irreversible — prefer dakera_memory_importance to suppress without deleting.
dakera_fulltext_searchBM25 keyword search over indexed documents. Use over vector search when exact-term recall matters (error codes, IDs, names). For semantic+keyword combined use dakera_hybrid_search.
dakera_hybrid_searchBM25 + vector ANN hybrid search in a single pass. Omit vector for BM25-only mode. Use for RAG when pure semantic or keyword search alone is insufficient. vector_weight: 0.0=BM25, 1.0=vector (default 0.5).
dakera_knowledge_graphBuild a knowledge graph from a seed memory using embedding similarity. Use to explore how a concept connects to stored knowledge. For BFS traversal of an existing linked graph use dakera_graph_traverse.
dakera_load_toolsFetch the full inputSchema for one or more named tools. Use after dakera_discover_tools. Returns schemas for found tools and a not_found list for unrecognized names.
dakera_recallRetrieve top-k memories semantically closest to a query. Prefer over dakera_batch_recall for query-based retrieval. Set include_associated=true to expand results via KG edges (1-3 hops).
dakera_searchSemantic search with optional tag and memory-type pre-filters. Prefer over dakera_recall when results must be constrained by tag or type alongside the semantic match.
dakera_session_endClose an active session with an optional summary. Always call at run end (even on error) to avoid orphaned sessions; summary is retrievable via dakera_session_get.
dakera_session_startOpen a new session, returning a session_id that groups stored memories under a shared context. Attach metadata such as task type or trigger source for later retrieval.
dakera_storePersist a new memory for an agent with importance weighting and optional tags. Use to save facts, decisions, or context for future retrieval. importance defaults to 0.5; set 0.8–1.0 for critical memories that must survive decay.
BEFORE YOU INSTALL
14 tools when we connected. 14 is the default core profile; DAKERA_MCP_PROFILE=admin gives 32, power 69 (68 in the config example) and all 87. dakera_discover_tools and dakera_load_tools pull in more on demand without switching profile.
All 4 documented setups run the server on your machine.
CONTEXT COST
Rated by the size of the 14 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
npm install -g @dakera-ai/dakera-mcp{
"mcpServers": {
"dakera": {
"command": "dakera-mcp",
"env": {
"DAKERA_API_URL": "http://localhost:3300",
"DAKERA_API_KEY": "<YOUR_DAKERA_API_KEY>"
}
}
}
}{
"mcpServers": {
"dakera": {
"command": "dakera-mcp",
"env": {
"DAKERA_API_URL": "http://localhost:3300",
"DAKERA_API_KEY": "<YOUR_DAKERA_API_KEY>",
"DAKERA_MCP_PROFILE": "power"
}
}
}
}MCP servers
Haustorium12
agentic-box
Obidel
sachitrafa