Live verified data for local AI agents.

Your local agent reasons on-device but its knowledge stops at training time. Point it at one keyless MCP URL and it gets 87 live tools - each datapoint stamped with its source, observed-at time and an Ed25519 signature it can verify. The same standard mcpServers config works across LM Studio, Jan, AnythingLLM, Msty and Cursor; Ollama-backed agents reach it through any MCP client.

The call

One config - the standard Cursor-style mcpServers block that LM Studio, Jan, AnythingLLM and Msty all accept - points a local agent at Dynamic Feed's keyless remote MCP:

{
  "mcpServers": {
    "dynamicfeed": { "url": "https://dynamicfeed.ai/mcp" }
  }
}

Sample response

# Where that config goes, per tool (all use the same mcpServers block):

LM Studio    → Program tab → Install > Edit mcp.json   (remote MCP since v0.3.17)
AnythingLLM  → storage/plugins/anythingllm_mcp_servers.json → Agent Skills → Refresh
Jan / Msty   → Settings → MCP Servers → add the remote URL above
Cursor / Claude / VS Code → same block (you may already have an mcpServers section)
Ollama       → Ollama serves the model; reach Dynamic Feed through an MCP-capable client
               (any of the above) or a framework (LangChain/CrewAI) pointed at /mcp.

# No MCP client at all? The keyless REST batch endpoint works anywhere:
curl -s -X POST https://dynamicfeed.ai/v1/batch \
  -H 'Content-Type: application/json' \
  -d '{"calls":[{"tool":"whats_changed_since_your_training","args":{}}]}'

Why live data

AI PCs with neural accelerators now ship from every major OEM, and the agents running on them use LOCAL LLMs - exactly when you want fresh facts instead of training-cutoff guesses. Dynamic Feed exposes its 87 tools over a keyless Streamable-HTTP MCP endpoint, so any MCP-capable local client loads the whole catalogue from one URL with no key. LM Studio follows Cursor's mcp.json notation and has supported remote MCP servers since v0.3.17; AnythingLLM reads an anythingllm_mcp_servers.json and hot-reloads from Agent Skills; Jan and Msty add the same remote URL in their MCP settings. Ollama itself is a model server, so an Ollama-backed agent reaches Dynamic Feed through one of those clients or a framework. Every result carries provenance (source, observed-at, freshness) and an Ed25519 signature the agent can verify - so a local model can fetch and CITE a current fact, and a consumer can check it wasn't altered. Be precise about the signature: it proves a datapoint existed unchanged at a time (tamper-evidence), not that the value is objectively true.

Use it for

• Give a local LM Studio / Jan / AnythingLLM agent 87 live tools from one keyless MCP URL
• Let an on-device model cite a current fact (CVE, software version, weather) instead of guessing
• Run an Ollama-backed agent (via LangChain/CrewAI) against the same keyless endpoint
• Verify each datapoint's Ed25519 signature locally — no trust in the pipe required

FAQ

Related live feeds