>_MCP.TOOLS

MCP Tool Surface

40+ tools for AI agents

AgentHALO exposes a comprehensive MCP tool surface for AI agents. NucleusDB operations, multi-agent orchestration, identity and trust management, container lifecycle, and Discord recording — all accessible via JSON-RPC over stdio or HTTP transports.

$ start mcp server

nucleusdb mcp --transport http --host 127.0.0.1 --port 3000 --db ./records.ndb

stdioHTTP (streamable)rmcp frameworkJSON-RPC 2.0

>_CATEGORIES

Tool Categories

NucleusDB

Orchestrator

HALO Identity & Trust

Container & Mesh

Discord

>_NUCLEUSDB

NucleusDB Tools

nucleusdb_get

Get a typed value by key

nucleusdb_set

Set a typed value with commitment

nucleusdb_delete

Delete a key (audit-logged in append-only mode)

nucleusdb_sql

Execute SQL queries against the verifiable core

nucleusdb_vector_search

kNN search over stored embeddings

nucleusdb_blob_store

Store content-addressed blobs

nucleusdb_blob_get

Retrieve blobs by SHA-256 hash

nucleusdb_memorize

Store agent memory with embeddings and tags

nucleusdb_recall

Retrieve memories via fused reranking

nucleusdb_chunk

Split content into indexed chunks

nucleusdb_reassemble

Reconstruct content from chunks

nucleusdb_work_record_*

Version-controlled work records

nucleusdb_verify

Verify commitment proofs for stored values

>_ORCHESTRATOR

Orchestrator Tools

orchestrator_launch

Launch a new agent in the pool

orchestrator_send_task

Send a task to a running agent

orchestrator_pipe

Pipe output between agents

orchestrator_graph

Define and query task dependency graph

orchestrator_mesh_status

View mesh network topology and health

orchestrator_stop

Gracefully stop a running agent

>_HALO.IDENTITY...TRUST

HALO Identity & Trust Tools

agenthalo_attest

Generate session attestation with Merkle proof

agenthalo_trust

Query trust scores and epistemic metrics

agenthalo_evidence_combine

Bayesian odds-update evidence combination

agenthalo_uncertainty_translate

Translate between probability/CF/possibility

deploy_preflight

Run preflight checks before agent launch

deploy_launch

Launch an agent from the deploy catalog

agenthalo_admission_check

Evaluate admission policy via governor

>_CONTAINER...MESH

Container & Mesh Tools

container_launch

Launch a containerized agent with Docker

container_logs

Stream container stdout/stderr

container_stop

Stop a running container

mesh_register

mesh_discover

Discover peers via DHT/gossip

mesh_tool_call

Cross-container MCP tool invocation

>_DISCORD

Discord Tools

discord_status

Bot and database status

discord_search

Search recorded messages

discord_verify

Verify a message record

discord_integrity

Full database integrity check

discord_export

Export records as JSON

>_MEMORY.PIPELINE

Memory Recall Pipeline

The memorize/recall tools use a sophisticated retrieval pipeline.

STORE

→Chunk text with overlap windows

→Generate 384-dim embeddings

→Store in NucleusDB (mem:chunk:* keys)

→Index in vector store

→Store blob (SHA-256 addressed)

RECALL

→Query expansion (optional LLM)

→Vector kNN search (4x candidate pool)

→Access-aware read (get_typed_touching)

→Fused rerank: similarity 0.50 + biencoder 0.28 + lexical 0.12 + negation 0.10

→Return top-k results (max 20)

Give Your Agent Full Access

One MCP connection unlocks verifiable storage, multi-agent orchestration, and trust attestation.

View on GitHub →Formal Verification