Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@crypto-insight-mcp What's the price of Bitcoin and Ethereum?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An MCP server that gives AI agents governed access to crypto market data and a company knowledge base — live prices and portfolio analytics from CoinGecko, plus RAG-based semantic search over internal documents (regulation, AML/KYC, custody, listing policy), with responsible-AI guardrails at every boundary.
Connecting an LLM to a financial domain is easy to do badly: unvalidated tool inputs, upstream stack traces leaking into model context, retrieved documents silently rewritten into confident "advice". This project demonstrates the architecture I consider correct for the problem:
One domain, two transports. All business logic lives in a pure service layer. An MCP server (stdio) exposes it to AI agents; a FastAPI gateway exposes the same functions to humans and systems. Neither transport contains logic, so behaviour and guardrails cannot drift between them.
Guardrails as a first-class module. Input validation with LLM-actionable
error messages, outbound rate limiting, mandatory not-financial-advice
disclaimers on every analytical response, structured {"error": ...}
payloads instead of exceptions crossing the protocol boundary.
Retrieval, not server-side synthesis. The RAG tool returns chunks with sources; the calling LLM does the reasoning. This division of labour is recorded in ADR-0003.
Runs anywhere, no keys. CoinGecko free tier, embedded Chroma, local ONNX
embeddings with a deterministic offline fallback. pytest passes with no
network at all (ADR-0002).
Related MCP server: crypto-projects-mcp
Tool | Arguments | Returns |
|
| Spot price + 24h change per symbol |
|
| Daily price points + min/max/change stats |
|
| Total value, per-position allocation %, HHI concentration index, warnings |
|
| Top-k knowledge-base chunks with |
Every analytical response includes a disclaimer field; every invalid input
produces {"error": "<what was wrong and what is acceptable>"} rather than a
crash.
flowchart LR
subgraph Agents
claude["Claude Desktop / MCP client"]
end
subgraph Humans["Humans & systems"]
rest["REST clients"]
end
claude -- "MCP (stdio)" --> srv["server.py\nFastMCP · 4 tools"]
rest -- "HTTP" --> api["api.py\nFastAPI gateway"]
srv --> svc["services.py\ndomain logic"]
api --> svc
svc --> guard["guardrails.py\nvalidation · rate limit · disclaimer"]
svc --> mkt["market/client.py\nTTL cache · token bucket"]
svc --> kb["rag/search.py\nKnowledgeBase"]
mkt -- "HTTPS" --> cg["CoinGecko free API"]
kb --> chroma[("Chroma embedded\n.chroma/")]
docs["knowledge_base/*.md"] -- "rag/ingest.py" --> chromaMore detail in docs/architecture.md and the ADRs.
Requires Python ≥ 3.10.
git clone https://github.com/IgorAbramov/crypto-insight-mcp.git
cd crypto-insight-mcp
pip install -e ".[dev]"
# Build the knowledge-base index (embedded Chroma, local embeddings).
python -m crypto_insight_mcp.rag.ingest
# Run the offline test suite.
pytestAdd to claude_desktop_config.json (Settings → Developer → Edit Config):
{
"mcpServers": {
"crypto-insight": {
"command": "crypto-insight-mcp",
"env": {
"CIM_CHROMA_DIR": "/absolute/path/to/crypto-insight-mcp/.chroma"
}
}
}
}If crypto-insight-mcp is not on Claude Desktop's PATH, use the absolute path
to the script (which crypto-insight-mcp) or
"command": "python", "args": ["-m", "crypto_insight_mcp.server"] with the
right interpreter. Restart Claude Desktop; then try:
What are BTC and ETH trading at? Then check what our listing policy says about delisting notice periods.
uvicorn crypto_insight_mcp.api:app --reload
# http://127.0.0.1:8000/docs — OpenAPI UI
# GET /health
# GET /prices?symbols=BTC,ETH&vs=usd
# POST /portfolio/analyze {"holdings": {"BTC": 0.5, "ETH": 10}}
# GET /knowledge/search?q=custody%20segregation&k=4Or with Docker:
docker compose up --build api # ingests on start, serves on :8000# Offline scripted mode — no LLM, no keys (needs internet for CoinGecko):
python agent_demo/demo.py "0.5 BTC, 10 ETH, 5000 USDT"
# Real tool-use loop through the Anthropic API:
pip install -e ".[agent]"
export ANTHROPIC_API_KEY=... # see .env.example
python agent_demo/demo.py "0.5 BTC, 10 ETH, 5000 USDT" --llmThe demo walks the agent workflow — prices → portfolio analysis → knowledge-base grounding → draft risk note — and then stops for human approval before "executing" the proposed action (execution is simulated; nothing is ever traded or sent).
Input validation at every tool boundary — symbols, query text, day ranges and holdings are validated and normalised; violations return messages that tell the LLM what was wrong and what acceptable values look like, so the agent can self-correct instead of retry-looping.
Rate limiting — a thread-safe token bucket in front of CoinGecko keeps a misbehaving agent from hammering a third-party API.
Mandatory disclaimers — every analytical payload carries
"Informational market data / document retrieval only. This is NOT financial, investment, legal or tax advice." The server's MCP instructions
direct clients to surface it.
No stack traces in model context — upstream failures map to short, safe
MarketDataError messages; tool handlers convert all handled errors to
structured {"error": ...} payloads, so the server never crashes on bad
input.
Human-in-the-loop — the agent demo requires explicit approval before any consequential action; the default answer is "no".
Retrieved chunks, not synthesized answers — search_knowledge returns
sourced chunks and leaves synthesis to the client LLM
(ADR-0003).
The suite runs fully offline: CoinGecko is mocked with
httpx.MockTransport, embeddings use a deterministic hash fallback, Chroma
lives in per-test temp directories, and the MCP surface is exercised
in-process (mcp.list_tools() / mcp.call_tool()).
pytest # 64 tests, ~1.5 s
ruff check . # lintCI (GitHub Actions) runs lint + tests on every push and pull request with no secrets configured — by design.
src/crypto_insight_mcp/
├── server.py # MCP transport (FastMCP, stdio)
├── api.py # REST transport (FastAPI)
├── services.py # domain logic shared by both
├── guardrails.py # validation, rate limiting, disclaimers
├── market/client.py # CoinGecko client: TTL cache, rate limit
└── rag/ # embeddings (ONNX + offline fallback), ingest, search
knowledge_base/ # sample corpus: MiCA, AML/KYC, custody, listing policy
agent_demo/demo.py # human-in-the-loop agent scenario (offline + --llm)
docs/ # architecture.md + ADRs
tests/ # offline test suitePinecone/managed vector-store adapter behind the existing LangChain interface (the embedded-Chroma trade-off is documented in ADR-0002).
Kubernetes manifests for the REST gateway.
Retrieval evaluation harness (golden questions → recall/precision on the knowledge base) to make RAG quality measurable, not anecdotal.
Symbol resolution fallback via CoinGecko /search for long-tail assets.
Igors Abramovs — github.com/IgorAbramov
MIT License — see LICENSE.
Value a portfolio and flag concentration risk (informational only).
Args: holdings: Mapping of ticker symbol to amount held, e.g. {"BTC": 0.5, "ETH": 10}. Max 50 positions. vs_currency: Quote currency (default "usd").
Returns total value, per-position allocation percentages, the HHI concentration index and warnings. No investment recommendations.
| Name | Required | Description | Default |
|---|---|---|---|
| holdings | Yes | ||
| vs_currency | No | usd |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided, so description carries full burden. Discloses 'informational only', 'No investment recommendations', max 50 positions constraint, and return fields (value, allocation, HHI, warnings). No contradictions.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Description is moderately concise and structured: purpose sentence, parameter list, return description, disclaimer. Could combine some lines, but all sentences add value.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given no output schema and moderate complexity (2 params, one nested), description covers purpose, parameters with constraints, returns, and usage caveats. Sufficient for an agent to invoke correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, but description adds significant meaning: explains holdings as 'Mapping of ticker symbol to amount held' with example and max 50 positions; clarifies vs_currency default. These details go beyond the raw schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description clearly states 'Value a portfolio and flag concentration risk', a specific verb+resource. Differentiates from sibling tools (get_price, get_market_history, search_knowledge) by focusing on portfolio analysis rather than price retrieval or market history.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Description implies use for portfolio valuation and concentration risk assessment. Does not explicitly state when not to use nor name alternatives, but the purpose is clearly distinct from siblings.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get daily price history for a symbol plus min/max/change statistics.
Args: symbol: Single ticker symbol, e.g. "BTC". days: Look-back window in days, 1..365 (default 30). vs_currency: Quote currency (default "usd").
| Name | Required | Description | Default |
|---|---|---|---|
| days | No | ||
| symbol | Yes | ||
| vs_currency | No | usd |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description should fully disclose behavioral traits. It mentions the output includes min/max/change statistics, but does not specify rate limits, error handling, or data freshness. For a read tool, more detail on safety and potential failures is needed.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is brief and front-loaded with the main purpose. The args section is clearly formatted with defaults and ranges. Every sentence adds value without redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Although the tool is simple and has no output schema, the description does not detail the return format (e.g., array of objects, fields). It covers the input parameters well but leaves the output structure underspecified for an AI agent.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, but the description clearly explains each parameter: symbol (single ticker), days (default 30, range 1-365), vs_currency (default 'usd'). This adds meaning beyond the empty schema descriptions.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states 'Get daily price history for a symbol plus min/max/change statistics,' clearly indicating the verb (Get) and resource (daily price history). It distinguishes itself from siblings like get_price (likely current price) and analyze_portfolio.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage for retrieving historical price data, but does not explicitly state when to use this tool versus alternatives like get_price or analyze_portfolio. No when-not or alternative guidance is provided.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get current spot price and 24h change for one or more ticker symbols.
Args: symbols: Ticker symbols, e.g. ["BTC", "ETH"]. Max 25 per call. vs_currency: Quote currency (default "usd").
| Name | Required | Description | Default |
|---|---|---|---|
| symbols | Yes | ||
| vs_currency | No | usd |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided, so description carries full burden. It discloses max 25 symbols per call and default quote currency, but lacks information on side effects, authorization needs, or return value details. Adequate but not rich.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is very concise with two short paragraphs. The first sentence front-loads the purpose, followed by parameter details. No wasted words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Without an output schema, the description does not specify the return format (e.g., fields like price, change%). It also omits data source or update frequency. For a price tool, more detail on output would improve completeness.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, but the description adds meaning by explaining symbols as ticker symbols with examples and a limit of 25, and vs_currency as quote currency with default 'usd'. This compensates for the missing schema descriptions.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description explicitly states it gets 'current spot price and 24h change' for ticker symbols, which is a specific verb and resource. It clearly distinguishes from siblings like get_market_history (historical data) and analyze_portfolio (portfolio analysis).
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies use when needing current price and 24h change, but does not explicitly state when not to use it or mention alternatives among siblings. Usage context is clear but no exclusions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Semantic search over the internal knowledge base (MiCA, AML/KYC, custody, listing).
Args: query: Natural-language question or keywords, max 500 characters. k: Number of chunks to retrieve, 1..10 (default 4).
Returns retrieved chunks with source and snippet — synthesise the
answer yourself and cite the sources.
| Name | Required | Description | Default |
|---|---|---|---|
| k | No | ||
| query | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Discloses that tool returns chunks with source and snippet, instructs agent to synthesize answer and cite sources. Specifies query length limit (500 chars) and k range (1-10). No annotations present, so description fully covers behavioral expectations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Concise, front-loaded with purpose, followed by parameter details and usage instruction. Every sentence adds value.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given simple parameter set and no output schema, description adequately covers input, behavior, and output format. No gaps for effective tool use.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Adds critical meaning beyond schema: query is natural language with max length, k is number of chunks with range. Since schema coverage is 0%, description provides all necessary parameter context.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Clearly states semantic search over internal knowledge base with specific topics (MiCA, AML/KYC, custody, listing). Distinct from sibling tools which handle price, market history, and portfolio analysis.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Implies usage for knowledge-based questions, and provides guidance on synthesizing answers and citing sources. However, no explicit when-not-to-use or alternative suggestions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
v0.1.0analyze_portfolioget_market_historyget_pricesearch_knowledgeEach tool targets a distinct function: price quotes, historical data, portfolio analysis, and knowledge retrieval. No overlapping purposes.
All tool names follow a consistent verb_noun snake_case pattern (get_price, get_market_history, analyze_portfolio, search_knowledge), making them predictable.
Four tools cover the core features of a crypto insight server without being excessive. Could potentially include one or two more for deeper analysis, but the count is reasonable.
The tool set covers common needs: price, history, portfolio valuation, and knowledge base. Minor gaps like market cap or news are absent but not critical for the stated purpose.
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
MCP server for querying Forkast documentation
MCP server giving AI agents one-connection access to crypto & DeFi data: DeFi protocol TVL, stableco
MCP server for Gainium — manage trading bots, deals, and balances via AI assistants
Hosted MCP server for live public-data APIs and Skills for AI agents.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/IgorAbramov/crypto-insight-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
