Skip to main content
Glama
positive-search-mcp

positive-search-mcp

#automation #ai #trading #n8n #prompt

Positive Search MCP server

Infrastructure for trading AI agents and their humans

CI Release Glama No dependencies Python 3.9+

which allows saving tokens while getting the highest-quality information about the current news background. Through the analysis of a large number of open sources, Polymarket, X and Truth.

An advanced scoring formula, a system of narratives and a source-quality audit together help not to let fakes through. And to easily make decisions based on information that is already analyzed, selected and sorted. The process of collecting the information and the process of analysis are laid out, described and easy to understand — and assembled from open sources, so, just in case, all of it is very easy to verify.

For the human on the site — a convenient UX where you immediately see which narratives pull the price and in which direction. It provides history in the blockchain, which gives very high reliability. The only tool to see the movement — sentiment over the last day, over the last few days, over the last week, which also helps to make better decisions in trading.

Want the details? How the AI reads the news and builds the index — About, in eight languages: EN · RU · DE · FR · ES · PT · PL · NL


Install

Python 3.9+. No dependencies — standard library only, so there is no version conflict to resolve and nothing to keep updated.

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "positive-search": {
      "command": "uvx",
      "args": [
        "--from", "git+https://github.com/IliaKostroma/positive-search-mcp",
        "positive-search-mcp"
      ]
    }
  }
}

Claude Code

claude mcp add positive-search -- uvx --from git+https://github.com/IliaKostroma/positive-search-mcp positive-search-mcp

Cursor

~/.cursor/mcp.json — same shape as the Claude Desktop block above.

Restart the client afterwards, then ask it "what's the news sentiment on bitcoin right now?"

Without uv

It is one file with no imports beyond the standard library, so a clone is enough:

git clone https://github.com/IliaKostroma/positive-search-mcp.git
python3 positive-search-mcp/ps_mcp.py --selftest

Then point command at python3 and args at the absolute path to ps_mcp.py.

Or install it into your environment:

pip install git+https://github.com/IliaKostroma/positive-search-mcp
positive-search-mcp --selftest

--selftest runs the handshake and every tool against the live site and prints a pass/fail line for each — worth running once before you trust it.


Related MCP server: Stelar Signals MCP

Tools

Tool

What it answers

get_sentiment

Start here. The current reading — number, state, measurement time and expiry, what it is made of (news + X + Polymarket), 1h/24h/7d change, a ready citation line — and the three heaviest narratives, so one call answers both what and why

get_narratives

The full set of stories, each with its sentiment, mass and cross-source agreement, plus example headlines with links

get_history

Seven days, daily by default (close/avg/min/max per day) — enough to tell a normal reading from an outlier; granularity: "hourly" for every run, about six times the payload

get_sources

The evidence — every headline that went into the reading, with its own score, age, publisher and a link to the original

compare_assets

All three readings at once, ranked most bullish first, each with its own measurement time

The cheap path: compare_assets to pick an asset → get_sentiment for the reading and the top narratives → get_narratives only if you want the whole picture → get_sources only to verify a specific headline → get_history only to check whether today is unusual.

Assets: BTC, GOLD, OIL. Loose spellings work — bitcoin, XAU, crude oil. Anything else returns an explicit error naming what is covered, never a substitute reading.

On weights. Narrative mass is published; per-article weight is not. A single article's weight is one model judgement and moves between runs, while narrative mass aggregates many articles and is stable — so weight is published where it holds still. Inputs are auditable, the published reading is tamper-evident, and the per-article weighting is not reproducible from here. We would rather say that than imply otherwise.


What makes it worth citing

Every reading carries its own measurement time. The three assets are scored on their own hourly runs, so their readings carry three different timestamps. Quote a number with the time next to it — a sentiment value without its moment is wrong the hour after.

The number is a stored fact, not a live recomputation. It is computed once, when the run happens, from the components captured in that same run. Re-read that run tomorrow and you get the same number to the last digit.

Every source is linked. The index is only as good as what it read, so what it read is public: get_sources returns each headline with its own score and a link to the original.

Past readings cannot be silently rewritten. Every run enters a sha256 hash chain whose head is anchored daily in the Bitcoin blockchain via OpenTimestamps. The fingerprints and their proofs are published at https://positivesearch.app/proofs/ and verify independently of this site and its owner (ots verify).

Honest limits, stated up front. It measures the tone of news coverage. It is not price, not a price forecast, and not financial advice. We do not recommend using it as your only input — use it alongside technical, fundamental and other analysis, as one instrument among several.


How to read the number

+1 strongly bullish coverage, −1 strongly bearish, 0 balanced, not silent — bull and bear pressure are measured separately, so a reading near zero usually means two strong narratives pulling against each other. get_narratives tells the two cases apart.

"Bullish" means works in favour of the price, not good news. Bad news for an industry is often bullish for its price.


Without MCP

Everything here is also plain HTTP, no key required:


Contact

Need another instrument (ETH, SOL, an FX pair, an index)? The engine is not tied to any asset — hi@positivesearch.app.

MIT licence for this client. The readings themselves are published under CC BY 4.0.

Available Tools

5 tools
compare_assetsA

All three readings (Bitcoin, Gold, Oil) in one call, ranked most bullish first, each with its own measurement time. Use when asked where the news mood is best or worst, or to pick an asset before going deeper with get_sentiment — three separate calls for the same picture is the most common waste.

Inputs: no named parameters.

get_historyA

Sentiment over the last seven days — reach for this only to tell a normal reading from an outlier, since today's number often only means something next to yesterday's. Daily by default: one point per day with close/avg/min/max, which answers that question at about a sixth of the payload. Pass granularity 'hourly' for every run only when the shape inside a day matters. Seven days is the full public window.

Inputs: asset, granularity.

get_narrativesA

The full set of stories behind the number — this is where most of the value is. Recurring narratives the hour's coverage groups into, each with its own sentiment, mass (its weight in the reading), cross-source agreement and example headlines with links. Use it to answer WHY sentiment is where it is, and to tell a quiet market (index near zero, no strong narratives) from a contested one (strong narratives pulling opposite ways). get_sentiment already returns the top three — call this when you want the whole picture or the headlines.

Inputs: asset, limit.

get_sentimentA

START HERE. Current AI news-sentiment reading for Bitcoin, Gold or crude Oil on a −1..+1 scale, rebuilt every hour from that hour's news coverage — plus the three heaviest narratives behind it, so one call answers both what the mood is and why. Also returns the measurement time and expiry, what the number is made of (news + X chatter + Polymarket odds), 1h/24h/7d changes, and a ready-to-use citation line. Covers BTC, GOLD and OIL only; any other asset returns an explicit error rather than a substitute. It measures the tone of coverage, not price, and does not forecast price.

Inputs: asset.

get_sourcesA

The evidence — reach for it to VERIFY, not to explain. Every headline that went into the current reading, each with its own score, age, publisher and a link to the original article. Use it to check a specific claim or to find the primary reporting behind a move; for the explanation itself, narratives are cheaper and clearer. Per-article weight is deliberately not published — weight is published where it is stable, as narrative mass.

Inputs: asset, limit.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.6/5.0

    Scored across 5 tools

    Disambiguation5/5

    Each tool targets a clearly distinct aspect: current sentiment, historical data, narrative details, source headlines, and cross-asset comparison. No two tools overlap in purpose, and descriptions explicitly differentiate when to use each, such as get_sentiment for the current reading vs. get_narratives for full stories.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern in snake_case: get_sentiment, get_history, get_narratives, get_sources, and compare_assets. The first four use 'get_' and the last uses 'compare_', which is still a verb_noun pattern, maintaining a predictable and uniform style.

    Tool Count5/5

    Five tools is well-scoped for this sentiment analysis server covering three assets. Each tool adds meaningful functionality without redundancy, and the count fits comfortably within the ideal 3-15 range for a focused API.

    Completeness5/5

    The tool surface covers all major aspects of sentiment analysis: current reading, historical trends, narrative breakdown, source evidence, and multi-asset comparison. Given the stated domain (news-sentiment for BTC, GOLD, OIL), there are no obvious gaps—the descriptions even mention users can follow up with more granular calls as needed.

    Maintenance

    ActivityStale
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      D
      maintenance
      Real-time cryptocurrency news, analysis, and price predictions for AI agents. 5 tools to search 50,000+ articles across 12 categories, filter by 120+ asset tickers, and access content with built-in attribution. Free with attribution. SSE and Streamable HTTP transport.
      4
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Real-time crypto news aggregation with sentiment scores and token mentions, enabling market analysis and trading signals via pay-per-call micropayments.
      MIT