neuroverse_assemble_context
Scan the codebase and assemble the most relevant file chunks based on a query.
LABSCO SUMMARY
The safety check sitting inside neuroverse_execute is the thing to look at closely, because that tool parses and runs a request end to end.
This is a platform rather than an integration, and the convenience tool that goes from a natural-language request to execution in one call is where its design decisions concentrate. Everything depends on what that safety check actually enforces, and the tool text says it exists without saying what it covers — worth reading the implementation before wiring it to anything consequential. The rest is a coherent set: tiered memory, model routing, local speech models rather than hosted ones, and HTTP routing to downstream agents. Local Whisper and Coqui is a real distinction for anyone who cannot send audio to a third party.
What it is
A broad agent-platform surface rather than an integration with one service: a processing pipeline, a tiered memory system, a model router, speech transcription and synthesis, codebase context assembly, downstream agent routing over HTTP, and feedback collection.
What you get
- neuroverse_store and neuroverse_recall work against a tiered memory system with short and longer-lived tiers, keyed per user.
- neuroverse_model queries a multi-model router, so the choice of model is made by the router rather than fixed at the call site.
- neuroverse_transcribe runs Whisper speech-to-text and neuroverse_synthesize runs Coqui text-to-speech, both local-model rather than hosted.
- neuroverse_route hands a task to a registered downstream agent over HTTP, which is what makes this a hub rather than a single agent.
- neuroverse_assemble_context scans a codebase and assembles the most relevant chunks for a query, and neuroverse_reason runs a complex task on higher-performance models.
- neuroverse_process runs mixed-language input through the whole pipeline, and neuroverse_execute parses, safety-checks and executes a request end to end.
- neuroverse_feedback submits RLHF data for tuning.
Requirements
The pipeline names Whisper and Coqui specifically, so speech depends on those models being available to it rather than on a hosted API.
Setup effort
One command — npm install neuroverse
USE CASES
01
Keeping context across sessions
Store records into the appropriate memory tier and recall them per user, rather than starting cold each time.
02
Working with audio without sending it away
Transcribe with local Whisper and synthesise with local Coqui, keeping the audio on infrastructure you control.
03
Letting the router pick the model
Query through the multi-model router, or escalate a hard task to the higher-performance reasoning path.
04
Handing work to another agent
Route the task to a registered downstream agent over HTTP rather than doing it in this one.
05
Giving a model the right part of a codebase
Assemble the most relevant file chunks for the query instead of shipping the whole tree.
06
Feeding corrections back in
Submit structured feedback for tuning, so a wrong answer becomes training signal rather than a retry.
11 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
neuroverse_execute
Parse, safety-check, and execute a user request end-to-end.
Convenience tool that chains: Language → Intent → Safety → Execute.
Args:
- text (string): Raw user input
- user_id (string): User / agent identifier
Returns: JSON with safety verdict and execution result
neuroverse_feedback
Submit Reinforcement Learning from Human Feedback (RLHF) data for agent tuning.
neuroverse_model
Query the multi-model AI router.
If a prompt is provided, the prompt is sent to the routed model. Otherwise, returns only the routing decision.
Supported providers: OpenAI, Anthropic, Sarvam AI, Ollama.
Args:
- task_type (string): multilingual | reasoning | local | general
- prompt (string, optional): Prompt to send
Returns: JSON with routing decision and optional model response
neuroverse_process
Process mixed-language input through the full NeuroVerse pipeline.
Pipeline: Language Detect → Normalise → Intent Extract → Safety Check → (optional) Execute
Supported languages: Tamil, Hindi, Telugu, Kannada + English (code-switched).
Args:
- text (string): Raw user input, possibly code-switched
- user_id (string): User / agent identifier (default: "anonymous")
- execute (boolean): Whether to also execute the intent (default: true)
Returns: JSON with keys: language, intent, safety, execution (if execute=true)
Examples:
- "anna indha file ah csv convert pannu" → detects Tamil+English, extracts convert_format
- "report banao sales ka" → detects Hindi+English, extracts generate_report
- "drop database production" → BLOCKED by safety layer
neuroverse_reason
Execute a complex reasoning task using specialized high-performance models (e.g. OpenRouter Reasoning). Returns the model's analytical response.
neuroverse_recall
Retrieve memories from NeuroVerse's tiered memory system.
Args:
- user_id (string): Agent / user identifier
- intent (string, optional): Filter by intent
- tier (string, optional): Filter by tier
- semantic_query (string, optional): Search constraint for vector engine
- limit (number): Max results (1–100, default 10)
Returns: JSON array of matching MemoryRecords
neuroverse_route
Route a task to a registered downstream agent via HTTP.
Args:
- target_agent (string): Name of the agent
- task (string): Task description
- payload (object): Arbitrary payload
Returns: JSON with the agent's response or a fallback error
neuroverse_store
Store a memory record in NeuroVerse's tiered memory system.
Tiers:
- short_term: In-process, capped at 50 per user. Lost on restart.
- episodic: Persisted to JSON file. Recent actions.
- semantic: Persisted to JSON file. Long-term facts.
Only episodic/semantic memories with importance_score ≥ 0.4 are persisted.
Args:
- user_id (string): Agent / user identifier
- intent (string): Canonical intent name
- tier (string): short_term | episodic | semantic
- language (string): Language code (default: "en")
- data (object): Structured payload
- importance_score (number): 0.0–1.0
Returns: JSON of the stored MemoryRecord
neuroverse_synthesize
Synthesize text to speech using Coqui TTS.
Args:
- text (string): Text to synthesize
- language (string): Language code
Returns: JSON with the path to the generated audio file
neuroverse_transcribe
Transcribe an audio file to text using Whisper STT.
Args:
- audio_path (string): Absolute path to the audio file
Returns: JSON with the transcribed text
BEFORE YOU INSTALL
What to know first
Tools you get
11 tools when we connected. The documented figure of 6 and its nine-row table are out of date; neuroverse_assemble_context and neuroverse_feedback are unlisted. The india_mcp_* names are the Python edition's old names, not extra tools.
Requires
- npx on PATH
- Node.js 18+
- Python edition only: Python 3.10+ and PostgreSQL
Where it runs
All 4 documented clients run the server on your machine.
CONTEXT COST
Light Standing cost
Rated by the size of the 11 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
Connect your client
Server setup
TERMINAL
npm install neuroverseMCP Clients
Cursor, Claude Desktop
JSON
{
"mcpServers": {
"neuroverse": {
"command": "npx",
"args": [
"neuroverse"
]
}
}
}VS Code
VS CODE · JSON
{
"servers": {
"neuroverse": {
"command": "npx",
"args": [
"neuroverse"
]
}
}
}Any MCP client — Python
ANY MCP CLIENT — PYTHON · JSON
{
"mcpServers": {
"neuroverse": {
"command": "python",
"args": [
"mcp/server.py"
],
"cwd": "<YOUR_PATH_TO_NEUROVERSE>"
}
}
}
