MCP Tools

The Sci MCP server exposes eight tools to connected agents.

memory_recall

Semantic search across episodic, semantic, and identity memory.

{
  "query": "what did I decide about auth last month?",
  "profile": "work",
  "limit": 10,
  "memory_types": ["episodic", "semantic", "identity"]
}

Parameter	Type	Default	Description
query	string	required	Natural language search query
profile	string	work	Profile to search (standard tier: always uses assigned profile)
limit	number	10	Max results
memory_types	array	all	episodic, semantic, identity

Returns results ranked by Reciprocal Rank Fusion across dense vector search and full-text search.

memory_store

Store a new episodic memory.

{
  "content": "Decided to use tsvector over Qdrant for simplicity of ops",
  "profile": "work",
  "source": "claude",
  "metadata": { "project": "sci" }
}

Parameter	Type	Default	Description
content	string	required	The memory to store
profile	string	work	Target profile
source	string	claude	Source agent or tool
metadata	object	{}	Optional metadata

Generates an embedding locally (no external API call) and persists to the storage backend.

memory_identity

Retrieve identity facts — preferences, values, skills, relationships.

{
  "query": "what stack does Casey prefer?",
  "category": "preference",
  "limit": 20
}

Parameter	Type	Default	Description
query	string	—	Semantic search query (omit for top facts by confidence)
category	string	—	Filter: preference, value, skill, relationship, project, context
limit	number	20	Max results

Auth restriction

Not available to standard tier agents. Identity facts are global (cross-profile).

memory_status

Health check.

{}

Returns backend name, row counts (episodic/semantic/identity/embeddings), and last write timestamp.

message_anonymize

Anonymize a message before sending to an AI provider.

{
  "text": "My name is Casey Zandbergen and I work on Threadline",
  "session_id": "optional-existing-session-id"
}

Returns anonymized text, a session_id for deanonymization, entities masked count, and a breakdown by type.

Use session_inspect to audit the token map before sending.

message_deanonymize

Restore real values in an AI response.

{
  "text": "Hello [PERSON_1], your project [PROJECT_2] looks good.",
  "session_id": "abc123",
  "discard_session": true
}

Set discard_session: true when the conversation is done to clear the token map from memory.

session_inspect

Audit the current token map for a session.

{
  "session_id": "abc123"
}

Returns every entity→token mapping. Use this to verify anonymization before any outbound call.

TIP

This is the launch gate: inspect the outbound text yourself and confirm your real name isn't there.

route_query

Get the recommended model for a given query.

{
  "query": "implement a TypeScript function that...",
  "context_tokens": 2000,
  "priority": "quality"
}

Parameter	Type	Default	Description
query	string	required	The query or task description
context_tokens	number	1000	Estimated input token count
priority	string	quality	quality, speed, or cost

Returns model, provider, reason, estimated cost, and context window.