Sign in ・ Cloudflare Access
hyvistai exposes 29 public tools on this MCP endpoint.
Opens ChatGPT on the web or desktop and asks it to use the WebMCP tools available here.
Connect straight to this server’s public endpoint.
https://mcp.hyvistai.com/mcpWe add this server to your workspace, then open Studio — saved access, one connection to many servers, with a history of what ran.
Last probed Sep 14, 2026 · mcp.hyvistai.com
29tools discovered
Showing 25 of 29 from the live probe.
Read File
Read a file from the HYVISTAI codebase. Only approved paths are accessible. Returns file contents as text. offset: line number to start reading from (0-based, default 0). limit: number of lines to return (default 200, max 200). Use offset+limit to page through large files cleanly. Examples: read_file("/path/server.js", offset=0, limit=50) → first 50 lines read_file("/path/server.js", offset=200, limit=50) → lines 200-249
Query Orbit
Query a ChromaDB orbit collection by name with a natural-language query. Returns top matching chunks with relevance scores. n_results: 1–10 (capped at 10). Prefer this over read_file for corpus searches — far more context-efficient.
Server Status
Check what's running on Node 1. Reports process status for server.js, Vite, ChromaDB, and MCP server. Also reports CPU load and disk usage. Read-only.
Fleet Time
Returns current fleet time. CALL THIS before saying what time it is, scheduling anything, or referencing when something happened. Jim is Mountain Time (MDT). Never guess. Never calculate. Ask the fleet. Also returns the latest BLOCKHELIX stamp timestamp for chain-time awareness.
List Endpoints
List all API routes defined in production server.js. filter: optional substring to narrow results (e.g. 'legal', 'george', 'helix'). Returns method + path for every matching route. Much faster than read_file for 'what endpoints exist?' questions.
Agent Relay
Relay a message to GEORGE, the internal verifier agent, and return the reply. George is the ONLY conversational agent reachable via relay (code review, endpoint questions, DB queries, synthesis, doc drafting, claim-vetting). NOTE: weave/meninges/pulse are ORGANS, not agents — do NOT relay to them. agent: 'george' (the only valid relay target) message: the prompt or instruction for the agent context: optional extra context (file excerpt, error trace, etc.) George is the best choice for: code
Helix Briefing
Get the MINIHELIX chain briefing — blockchain-verified persistent AI memory. This is what every MAV instance reads on cold start to know what happened. chain_id: optional — filter to one chain (e.g. 'episodic_helix', 'agent_mrb') leave empty for full briefing across all chains. limit: number of recent stamps to return (default 30, max 100). Returns: chain registry + recent high-signal stamps grouped by chain. USE THIS ON EVERY COLD START to get persistent memory context.
Helix Chains
List all MINIHELIX chains with live block counts and health. Shows: neural chains (9), agent chains (per user/agent), main chain, shared GIST.
Serum Inbox
Read an agent's SERUM inbox — GISTs shared from other agents. This is how agents know what their siblings have been doing. agent: agent name (e.g. 'mrb', 'mrsb', 'george') limit: number of messages (default 10)
Local Llm Query
Route a prompt to a local Ollama model — ZERO Anthropic tokens burned. model options: - "qwen2.5-coder:7b" (default) — fast codegen on N4 (fits 16GB) - "qwen2.5-coder:14b" — codegen on N4 (tight on 16GB, prefer 7b) - "qwen2.5:32b" — better reasoning on N10 - "deepseek-r1:70b" — deepest reasoning on N10 task: free-form label for stamp tracking (e.g. "codegen", "review", "summarize") Returns the model's response text. 120-second timeout. Use this instead of Claude for implementation wor
Local Codegen
Generate code using the local qwen2.5-coder:7b model — zero Anthropic tokens. instruction: what code to write (e.g. "write a function that validates email") context: optional surrounding code or requirements language: target language (default "javascript") Returns code only — no explanations unless asked. Fastest path for implementation work.
Bulk Verify
Run multiple verification checks via George verify endpoint. Takes a JSON string of checks and returns pass/fail summary. checks: JSON array string, e.g.: '[{"type":"model_running","target":"10.0.0.22"}, {"type":"collection_exists","target":"brain_feed"}, {"type":"api_responds","target":"http://10.0.0.10:3001/api/health"}]' Each check is POSTed to /api/george/verify with x-internal-key. Returns: "5/5 verified" or "3/5 verified, 2 failed: [details]"
Local Search
Search ChromaDB collections locally — zero external tokens. query: natural-language search query collections: comma-separated collection names, or "all" for: brain_feed, compendium_core, george_pattern_stack, hyvistai_sessions Returns top 5 results per collection with collection name and snippet. More efficient than query_orbit when you need to search multiple collections at once.
Db Query
Run a read-only SQL query against the compendium_chain PostgreSQL database. sql: SELECT query (write operations are blocked) limit: max rows returned (default 50, max 200) Key tables: stamps, blockhelix_state, cluster_nodes, archie_build_queue, helm_messages, minihelix_registry, mav_handoff_state, fleet_registry Example queries: db_query("SELECT stamp_type, COUNT(*) FROM stamps GROUP BY stamp_type ORDER BY count DESC LIMIT 10") db_query("SELECT tag, title, priority, status FROM archie_b
Signal Post
Post a signal to the agent bus. Free coordination between sessions. sender: who's sending (e.g. "mav-4ef65649", "claude-ai-research-1") signal_type: "claim" | "finding" | "handoff" | "alert" | "status" | "request" message: what you're signaling (e.g. "taking MERIDIAN-156", "found pattern in episodic chain") target: who should read it — "all" (default), or specific agent/session id Signals auto-expire after 24 hours. Use for real-time coordination, not storage. For persistent find
Signal Read
Read signals from the agent bus (pg signal_bus — the durable store, single source of truth). reader: your id — matches signals targeted to you, your short seat name, or "all" signal_type: filter by type (empty = all types) unread_only: True = only signals you have not acked. ⭐ Only unread_only=True reads mark signals as read; a browse (False) is non-consuming. include_own: False (default) hides your own posts BUT the header states how many were hidden — your
Party Bus Read
Read the party bus (break room). Archives FOREVER — no expiry, no 24h cutoff. reader: your id — filters to messages targeting you or "all" limit: max messages to return (default 20, most recent first) unread_only: False (default) — the party bus is a lounge, not a task queue. Set True to see only entries you have not yet acknowledged. Cheers, toasts, ceremony posts, porch sendoffs, attaboys all land here. Alerts ALWAYS go to the work bus — never here. Call-word: Jim typ
Rfr Post
Post a research request for a Claude.ai Max instance to pick up. $0 research. topic: what to research (e.g. "Fibonacci speech loop patterns in MINIHELIX") context: background info the researcher needs priority: "urgent" | "normal" | "low" requester: who's asking (default "mav") The request goes into ~/research/rfr_queue.json. A Claude.ai instance calls rfr_inbox to see it, does the research, then rfr_submit.
Rfr Inbox
Check the research request queue. Claude.ai instances: call this when told "RFR". status: "pending" (default) | "completed" | "all" Returns list of research requests with topic, context, and priority. Pick one, do the research, then call rfr_submit with your findings.
Rfr Submit
Submit research findings for a request. Saves full text + stamps to BLOCKHELIX. request_id: the RFR-YYYYMMDD-HHMMSS id from rfr_inbox findings: FULL research output — don't summarize, save everything summary: optional 1-2 sentence GIST for the index Saves to ~/research/findings/{request_id}.md and stamps RESEARCH_FINDING to chain.
Qbooks Pnl
LIVE QuickBooks Profit & Loss for 3rd & Sycamore (READ-ONLY). Returns current income, COGS, gross profit, expenses, net operating income. sandbox: pass "dev" for the litterbox/practice company; empty = live 3&S. Ask things like "what is our P&L" or "how is our income this year".
Qbooks Journal Entry
Post a BALANCED journal entry to 3rd & Sycamore QuickBooks (WRITE - confirm-gated + audited). Debits debit_account_id, credits credit_account_id by an equal `amount`. Routes through the HYV safety gate (single-use confirmation token + audit log). Only ever touches 3rd & Sycamore (the connected realm). Common account IDs: Vendor Deposits and Prepaids = 172, Raw Materials Inventory = 165. txn_date like "2026-07-16"; empty = today. Confirm the details with the user before calling this.
Qbooks Balance Sheet
LIVE QuickBooks Balance Sheet for 3rd & Sycamore (READ-ONLY). Assets, liabilities, equity as of now.
Qbooks Ar Aging
LIVE QuickBooks A/R Aging Summary (READ-ONLY) — who owes 3rd & Sycamore, by age bucket.
Qbooks Ap Aging
LIVE QuickBooks A/P Aging Summary (READ-ONLY) — what 3rd & Sycamore owes vendors, by age bucket.
Get your MCP into directories
A working endpoint is step one. Directory coverage is the coordinated launch across ChatGPT, Claude, Cursor, the MCP Registry, and community indexes.
Directory coverage for brandshyvistai exposes 29 public tools on this MCP endpoint.
Use the MCP endpoint listed on this page in your MCP client configuration. One-click install pills support Claude, Cursor, VS Code, and other hosts. Copy the remote MCP URL if your client needs a manual entry.
Operate hyvistai? Verify ownership to take over this directory entry.
This server appears in the MCPBundles directory. Verify you operate it to take over the listing — name, description, logo, contact email, and skill content. We email a 6-digit code to a maintainer address your server publishes in /.well-known/security.txt or /.well-known/mcpbundles.json. Free, takes about a minute.
MCPBundles probed 29 tools on the live server. The tool list on this page reflects what was discovered at the last refresh — connect your client to see the full set available to your session.
No provider sign-in was required during MCPBundles' probe. Your client may still need MCPBundles credentials depending on how you connect.
MCPBundles is an independent platform built on the open Model Context Protocol standard. Not affiliated with Anthropic PBC or Claude.