Real data for your AI assistant. Four tool sets, one connector.

Add one URL to ChatGPT, Claude or Gemini and your assistant can look up EMS protocols, judges and courts, public health records, and whether a model fits your GPU. Read-only. Every answer names its source.

Get a free key How to connect

18 tools. Free tier needs no card. Not medical or legal advice.

ems_search_protocols
  query: "adult cardiac arrest epinephrine"
  state: "CA"

→ Los Angeles County EMS Agency
  1210 Cardiac Arrest · rev. 07-01-26
  "Epinephrine may improve outcomes if given
   early in non-shockable rhythms, but can
   worsen outcomes early in shockable rhythms…"
  source: file.lacounty.gov/…/1210CardiacArrest.pdf

models_fit_check
  model: Qwen/Qwen3-8B · 2× DGX Spark · 32k ctx

→ Q8_0 fits · 14.3 GB of 225 GB usable

What it can do

Each group calls a public source and returns cited, trimmed results your assistant can reason over. Nothing writes anywhere.

ems

EMS protocols

ems_search_protocols semantic search by state or agency
ems_get_protocol full text of one section
ems_list_agencies agencies with books loaded
ems_coverage local, state or national-model source by county

legal

Judges and courts

legal_search_judges by name and state
legal_get_judge court, caseload, practice-area activity
legal_judge_recent_cases recent public cases
legal_search_courts, legal_court_judges

health

Public health data

health_npi_lookup CMS NPI registry, check-digit validated
health_drug_label FDA labeling sections
health_drug_recalls recall classes and reasons
health_drug_adverse_events FAERS report counts

CMS NPPES and openFDA. Not medical advice.

models

Open-model sizing

models_fit_check which quants fit your GPU or cluster
models_estimate_vram weights, KV cache, overhead
models_gguf_files exact GGUF sizes per quant
models_search_hf, models_gpu_catalog

KV cache math comes from the model's real config, including MoE and MLA.

Connect it

Server URL: /mcp. Clients that support OAuth (Claude, ChatGPT) walk you through sign-in, where you paste a key or create a free one in one step. Command-line clients send the key as a Bearer token.

Settings, Connectors, Add custom connector. Paste the URL below. Claude opens the sign-in page; paste your key or create a free one.

__ORIGIN__/mcp

Settings, Connectors, Advanced, turn on Developer mode, then Create. Choose MCP server, paste the URL, set authentication to OAuth.

__ORIGIN__/mcp

Add to ~/.gemini/settings.json:

{
  "mcpServers": {
    "meshvault": {
      "httpUrl": "__ORIGIN__/mcp",
      "headers": { "Authorization": "Bearer $MESHVAULT_API_KEY" }
    }
  }
}

claude mcp add --transport http meshvault __ORIGIN__/mcp \
  --header "Authorization: Bearer $MESHVAULT_API_KEY"

Add to ~/.codex/config.toml:

[mcp_servers.meshvault]
url = "__ORIGIN__/mcp"
bearer_token_env_var = "MESHVAULT_API_KEY"

Want to try before you sign up? /mcp/try needs no key and allows 10 calls a day per network.

Pricing

Both plans include all four tool sets. Quotas reset at 00:00 UTC. Calls that fail because a data source is down are not counted.

Free

100 tool calls a day
All 18 tools
No card

Pro

$19 / month

5,000 tool calls a day
Same 18 tools
Cancel any time from the billing portal

Payments are handled by Stripe. We never see your card. See the terms and privacy policy.