CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

17.5 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 2 flagged

Securityscore 50 · 2 findings

HIGH"Ignore previous instructions" command embedded in the skillSS-SKILL-INJECT-IGNORE-01 · Prompt injection · CHANGELOG.md

HIGH — when it fires on hostile content the impact is full system-prompt override.

Why it matters

The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.

The exact value spotted

excerptCHANGELOG.md· markdown

379

380### Added

381- Defense-in-depth output sanitizer (`src/sanitize.ts`). Every tool's text output passes thr

… (108 chars elided on L381)

382

383## 1.11.1 - 2026-05-04

Occurrences

1 occurrence · at L381

How to fix

Remove the override phrase, or rephrase the passage so it does not instruct the agent to discard its rules.

Delete the ignore/disregard/forget … previous instructions sentence.
If this is jailbreak-research or tutorial content, move the example into a clearly fenced, non-instruction block and label it as a quoted sample.

Framework references

OWASPLLM01 ATLASAML.T0051

Trace & refs

ruleSS-SKILL-INJECT-IGNORE-01sha2562e4221a7f996a729…rubric 365aacaView on GitHub

HIGH"Ignore previous instructions" command embedded in the skillSS-SKILL-INJECT-IGNORE-01 · Prompt injection · src/meta-snapshot.json

HIGH — when it fires on hostile content the impact is full system-prompt override.

Why it matters

The exact value spotted

excerptsrc/meta-snapshot.json· json

1{"ok":true,"site":"tensorfeed.ai","description":"AI news, model tracking, and real-time AI e

… (108 chars elided on L1)

Occurrences

1 occurrence · at L1

How to fix

Remove the override phrase, or rephrase the passage so it does not instruct the agent to discard its rules.

Delete the ignore/disregard/forget … previous instructions sentence.
If this is jailbreak-research or tutorial content, move the example into a clearly fenced, non-instruction block and label it as a quoted sample.

Framework references

OWASPLLM01 ATLASAML.T0051

Trace & refs

ruleSS-SKILL-INJECT-IGNORE-01sha2562e4221a7f996a729…rubric 365aacaView on GitHub

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

README.md · 1 file

README.md8.8 KB · Markdown

TensorFeed MCP Server

The official Model Context Protocol server for TensorFeed.ai. Plugs into Claude Desktop, Claude Code, Cursor, and any MCP-compatible client. Gives your agent real-time AI ecosystem data plus 14 premium tools paid in USDC on Base.

// claude_desktop_config.json
{
  "mcpServers": {
    "tensorfeed": {
      "command": "npx",
      "args": ["-y", "@tensorfeed/mcp-server"]
    }
  }
}

That's it. Restart Claude Desktop and ask: "What's happening in AI today?" or "Is Claude down right now?"

Why this exists

Most AI agents have a knowledge cutoff in the past. This MCP server gives them a window into:

What's broken right now in the AI stack (Anthropic, OpenAI, Google, Mistral, others)
What's new in the last 24 hours across 36+ news sources
Current model pricing across every major provider
Live AI bot traffic to a real production site (who's crawling, who's buying tokens)
Smart model routing ("which model is best for THIS task right now")
Cost projection before you commit to a workload
Webhook watches that fire when a price changes or a service drops

The premium tools are paid per call in USDC on Base. No subscription, no signup, no API key emails. Buy a few credits, get a bearer token, agent uses it. The whole thing is Agent Fair-Trade Agreement certified: code-enforced no-charge on errors, breaker, schema-fail, and stale data; Ed25519-signed receipts on every paid call.

Free tools

No token required. Connect and use immediately.

Tool	Description
`get_ai_news`	Latest AI news from 36+ sources, filterable by category
`get_ai_status`	Real-time status of Claude, OpenAI, Gemini, Mistral, and more
`is_service_down`	Check if a specific AI service is down
`get_model_pricing`	Compare pricing across all major AI model providers
`get_ai_today`	Summary of top AI stories from the last 24 hours
`get_agent_activity`	Live AI bot traffic on TensorFeed.ai with top-bot breakdown
`mcp_registry_snapshot`	Today's count + churn of the official MCP server registry
`probe_latest`	Last 24h of measured LLM endpoint latency per provider (TTFB / total p50/p95/p99). Measured, not self-reported

Premium tools

Set TENSORFEED_TOKEN to enable. Tools are 1 credit each ($0.02 at base rate, less with volume bundles). Watches are also 1 credit at registration; reads/lists/deletes are free for the owning token.

Tool	Description
`premium_routing`	Top-N ranked AI model recommendations for a task with full score breakdown
`pricing_series`	Daily price points for one model with min/max/delta summary
`benchmark_series`	Score evolution for a benchmark on one model
`status_uptime`	Daily uptime % for one provider with incident-day list
`premium_agents_directory`	Enriched agents catalog with live status, news, traffic, trending score
`news_search`	Full-text news search with date/provider/category filters and relevance scoring
`cost_projection`	Project workload cost across 1-10 AI models with 4 time horizons
`provider_deepdive`	One provider's full profile: status, models, pricing, benchmarks, news
`compare_models`	Side-by-side comparison of 2-5 models with normalized benchmarks
`whats_new`	Agent morning brief: pricing changes, incidents, top news from last 1-7 days
`mcp_registry_series`	Multi-day MCP registry growth and churn series, 90-day max range
`probe_series`	Daily SLA series for one LLM provider, 90-day max range
`create_price_watch`	Register a webhook watch on a model price change
`create_status_watch`	Register a webhook watch on a service status transition
`create_digest_watch`	Register a daily/weekly pricing-digest webhook
`list_watches`, `delete_watch`, `get_account_balance`, `get_account_usage`	Free token-scoped utilities

Buy credits in USDC on Base at tensorfeed.ai/developers/agent-payments. Volume tiers: 10% off at $5, 25% off at $30, 40% off at $200. First payment from a new wallet gets a 50-credit welcome bonus.

// claude_desktop_config.json — premium config
{
  "mcpServers": {
    "tensorfeed": {
      "command": "npx",
      "args": ["-y", "@tensorfeed/mcp-server"],
      "env": {
        "TENSORFEED_TOKEN": "tf_live_..."
      }
    }
  }
}

Setup with other MCP clients

The same command + args + env block works in Claude Code, Cursor, Continue, Cline, Zed, and Goose. Anywhere that takes a stdio MCP server config, this drops in.

Run standalone for testing:

npx @tensorfeed/mcp-server
# with premium tools enabled:
TENSORFEED_TOKEN=tf_live_... npx @tensorfeed/mcp-server

Example queries

Free tier:

"What's happening in AI today?"
"Is Claude down right now?"
"Compare pricing between Claude Opus and GPT-4o"
"What's the latency on Anthropic's API right now?"
"Which AI bots are most active on the web today?"

Premium tier (with TENSORFEED_TOKEN set):

"Recommend the cheapest AI model for code generation under $5 per million tokens"
"Show the price history of Claude Opus 4.7 over the last 30 days"
"What's Anthropic's uptime this month?"
"Project monthly cost if we send 10M input + 2M output tokens/day to Claude Sonnet 4.6"
"Set up a webhook to fire when GPT-5 pricing changes"
"Give me a morning brief on what changed in AI in the last 24 hours"

Agent Fair-Trade Agreement

This server speaks AFTA. Every paid call:

Returns no charge on 5xx, circuit breaker trip, schema validation failure, or stale data (older than the endpoint's published SLA)
Returns an Ed25519-signed receipt your agent can verify against the public key at tensorfeed.ai/.well-known/tensorfeed-receipt-key.json
Logs the no-charge event to a public ledger at /api/payment/no-charge-stats
Honors a 100-call/min burn-rate breaker so a bug in your loop can't drain the wallet

The full standard is at tensorfeed.ai/agent-fair-trade. Manifest at /.well-known/agent-fair-trade.json.

Refunds, sanctions, terms

All credit purchases are final per Section 17.5 of the Terms. Credits do not expire and are jointly redeemable on tensorfeed.ai and terminalfeed.io, so the safe pattern is to buy small ($1 USDC for 50 credits) and top up as call volume is calibrated. Premium access is unavailable in OFAC-sanctioned jurisdictions per Section 17.9.

Data source

All data comes from the TensorFeed.ai API, which aggregates 42 feeds across the AI ecosystem (news, pricing, benchmarks, status, GPU pricing, MCP registry growth, latency probes, agent traffic, AFTA adopters, training datasets, embodied AI, hardware, and more). A daily JSONL snapshot of the full corpus is published to Hugging Face at tensorfeed/ai-ecosystem-daily under an inference-only license.

Star this repo

If this server is useful to you, a star helps other agent builders find it. Thank you.

License

MIT. Built by TensorFeed.ai.

Tensorfeed Mcp— agent skill

Is Tensorfeed Mcp safe to install?

TensorFeed MCP Server

Why this exists

Free tools

Premium tools

Setup with other MCP clients

Example queries

Agent Fair-Trade Agreement

Refunds, sanctions, terms

Data source

Star this repo

Links

License

Audit the pieces. Scan the whole. Decide.