MCP Server

MIKO is exposed as a Model Context Protocol server. Any MCP-aware client can call MIKO's capabilities as tools directly inside its own conversation context, so a holder can use MIKO from Claude Desktop, Cursor, OpenAI Agents, Gemini, or any other MCP-aware workflow.

What is MCP

MCP (Model Context Protocol) is an open protocol for connecting AI applications to external data sources and tools. An MCP server exposes a set of tools that an LLM-host client can discover and invoke during conversation.

Server Package

@project/mcp-server (npm)

The server runs locally via npx. It reads the holder's wallet JWT from an environment variable and connects to the MIKO API.

Authentication

The MCP server requires a JWT obtained through the REST API authentication flow (see REST API → Authentication).

The JWT is passed via environment variable:

MIKO_JWT=<wallet_signed_jwt>

The JWT can be obtained either through the REST /auth/nonce + /auth/verify/ flow ro by signing in at https://api.mikoprotocol.com/dashboard and copying the bearer toekn.

Tier and quota are evaluated at every tool call.

Tool Access by Tier

Tool	Minimum Tier
`miko.factcheck`	Pro
`miko.narrative`	Pro
`miko.persona`	Pro
`miko.watchlist`	Pro
`miko.insights`	Holder
`miko.narratives_trending`	Holder

Calls to a tool above the caller's tier return 403 tier_insufficient.

Tools

miko.factcheck

Verify a claim through MIKO's multi-provider fact-check.

{
  "name": "miko.factcheck",
  "description": "Verify a factual claim through MIKO's multi-provider fact-check.",
  "input_schema": {
    "type": "object",
    "properties": {
      "claim": { "type": "string", "description": "The claim to verify" }
    },
    "required": ["claim"]
  }
}

In an agent chain: the verdict's tri-state (true / false / null) typically gates the next tool call — for example, abort code generation that depends on a false claim, or surface the reasoning text in the final answer.

miko.narrative

Get MIKO's narrative read of a Solana mint address: market snapshot, one-paragraph interpretation, and concrete observations.

{
  "name": "miko.narrative",
  "description": "Get MIKO's narrative read of a Solana token: what is the token doing on-chain right now, expressed as a one-paragraph interpretation plus a list of concrete observations.",
  "input_schema": {
    "type": "object",
    "properties": {
      "mint_address": { "type": "string", "description": "Solana mint address" }
    },
    "required": ["mint_address"]
  }
}

In an agent chain: the agent reads narrative_summary and observations to brief the user before suggesting any action, or to enrich subsequent on-chain queries before recommending a swap.

miko.insights

Get MIKO's insights on a token or narrative topic.

{
  "name": "miko.insights",
  "description": "Get MIKO's insights on a token symbol, mint address, or narrative topic.",
  "input_schema": {
    "type": "object",
    "properties": {
      "query": { "type": "string", "description": "Token symbol, mint address, or narrative topic" }
    },
    "required": ["query"]
  }
}

In an agent chain: the agent picks verified items as grounding context for its own reasoning, citing last_confirmed_at timestamps when responding to the user.

miko.narratives_trending

Top narratives in the requested window, ranked by importance and uniqueness.

{
  "name": "miko.narratives_trending",
  "description": "Get the top narratives in the requested window, ranked by importance and uniqueness. Returns each narrative's summary, key entities, key token symbols, observation timestamps, momentum direction, and fact-check status.",
  "input_schema": {
    "type": "object",
    "properties": {
      "window": {
        "type": "string",
        "enum": ["1d", "2d", "3d", "4d", "5d", "6d", "7d"],
        "default": "1d",
        "description": "Aggregation window"
      }
    }
  }
}

In an agent chain: the agent skips downstream tool calls on topics whose momentum is cooling, or batches further research on rising themes.

miko.watchlist

Tokens currently on the watchlist, each with a plain-language summary.

{
  "name": "miko.watchlist",
  "description": "Tokens currently on the watchlist, each with a plain-language summary.",
  "input_schema": {
    "type": "object",
    "properties": {}
  }
}

In an agent chain: the agent uses the watchlist as a starter set for deeper per-token research (miko.narrative, miko.insights) before producing its final answer.

miko.persona

Generate a response in Miko's voice.

{
  "name": "miko.persona",
  "description": "Generate a short response in Miko's voice. Pass the content to react to or the topic to write about; the persona is built into the model.",
  "input_schema": {
    "type": "object",
    "properties": {
      "prompt": { "type": "string", "description": "What Miko should respond to or write about" }
    },
    "required": ["prompt"]
  }
}

In an agent chain: the agent uses miko.persona to phrase its final answer in Miko's voice, or to draft an in-character reply for a downstream channel. See Open Model for the same model run locally or over the REST API.

Connecting from Claude Desktop

Claude Desktop reads MCP server configurations from a JSON file.

Location:

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

Add a miko entry under mcpServers:

{
  "mcpServers": {
    "miko": {
      "command": "npx",
      "args": ["-y", "@projectmiko/mcp-server"],
      "env": {
        "MIKO_JWT": "<your_wallet_jwt>"
      }
    }
  }
}

Save the file and restart Claude Desktop. The MIKO tools appear in the tools panel of any new conversation.

Conversation example

User: I saw a claim on X that $BONK is partnering with Visa next week.
      Worth tracking, or just noise?

Claude: I'll verify this through MIKO.
        [Tool call: miko.factcheck({"claim": "$BONK partnering with Visa next week"})]

        MIKO returned NOT verified:
        - No Visa press release found
        - No official $BONK channel has confirmed
        - The original post traces to an unverified account with a
          history of similar pump claims

        This is noise. Treat it as unverified until a primary source surfaces.

User: What's $BONK actually doing on-chain right now?

Claude: [Tool call: miko.narrative({"mint_address": "DezXAZ8z7..."})]

        MIKO's read:
        - Accumulation phase with steady buy-side pressure but limited
          breakout participation
        - 24h volume up roughly 40% versus the prior 24h
        - Trade size declining while count rises — distributed
          accumulation pattern rather than retail-driven move
        - Price holding above the 4h high with shallow pullbacks

        Solid structure, but not yet breakout territory.

User: Is BONK worth a position right now given the broader market?

Claude: [Tool call: miko.narratives_trending({"window": "7d"})]
        [Tool call: miko.insights({"query": "BONK"})]

        Narrative context this week is dominated by AI agent infrastructure
        (CLAWD, MOLT), with memecoin discussion cooling. BONK sits in the
        cooling part of the rotation.

        MIKO's insights on BONK include a verified entry on community
        staking activity from Q1 2026 and a longstanding observation of
        repeated unverified exchange-listing claims that never
        materialised.

        So: structurally healthy on-chain, but narrative tailwind is rotating
        away from memecoins this week. Position sizing should reflect that.

Connecting from Cursor

Open Cursor Settings → Features → Model Context Protocol. Add:

{
  "mcpServers": {
    "miko": {
      "command": "npx",
      "args": ["-y", "@projectmiko/mcp-server"],
      "env": {
        "MIKO_JWT": "<your_wallet_jwt>"
      }
    }
  }
}

The MIKO tools appear in Cursor's chat panel. Useful for in-editor crypto context lookups while writing code that touches Solana.

Usage example

Cursor user: I'm writing a Jupiter swap script for $BONK. Quick sanity check —
             is $BONK currently in good liquidity for a $5K swap?

Cursor: [Tool call: miko.narrative({"mint_address": "..."})]

        BONK 24h volume is $85M, market cap $1.24B. A $5K swap is roughly
        0.006% of 24h volume. Liquidity is more than sufficient for that
        size. MIKO's read: accumulation phase, no concentration warnings.

Connecting from OpenAI Codex / Agents SDK

OpenAI's Agents SDK has native MCP support.

from openai import OpenAI
from openai.agents import Agent, MCPServer

agent = Agent(
    model="gpt-5",
    instructions=(
        "You are a Solana research assistant. Use MIKO tools to verify "
        "claims and read tokens. Always run miko.factcheck before reporting "
        "any claim as fact."
    ),
    tools=[
        MCPServer(
            command="npx",
            args=["-y", "@projectmiko/mcp-server"],
            env={"MIKO_JWT": "<your_wallet_jwt>"}
        )
    ]
)

result = agent.run("""
Verify this claim and tell me if the underlying token is worth tracking:
'$BONK is announcing a Visa partnership next week, this is huge'
""")
print(result.output)

The agent autonomously decides when to call miko.factcheck and miko.narrative based on the user's query.

Connecting from Gemini

Gemini's function calling system can be bridged to MCP via the MCP Python client.

import google.generativeai as genai
from mcp_client import MCPSession

genai.configure(api_key="<gemini_api_key>")

# Spin up MCP session connected to MIKO
mcp = MCPSession.connect_stdio(
    command="npx",
    args=["-y", "@projectmiko/mcp-server"],
    env={"MIKO_JWT": "<your_wallet_jwt>"}
)

# Convert MCP tools to Gemini's function declaration format
miko_tools = mcp.tools_to_gemini()

model = genai.GenerativeModel(
    "gemini-2.5-pro",
    tools=miko_tools,
    system_instruction=(
        "You are a Solana research assistant. Verify claims through "
        "miko.factcheck before treating them as fact."
    )
)

chat = model.start_chat()
response = chat.send_message(
    "Read $BONK and tell me if it's worth tracking right now"
)
print(response.text)

Connecting from Custom Agents

Any agent runtime that speaks MCP over stdio can connect using the official Python MCP client:

from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client

server_params = StdioServerParameters(
    command="npx",
    args=["-y", "@projectmiko/mcp-server"],
    env={"MIKO_JWT": "<your_wallet_jwt>"}
)

async with stdio_client(server_params) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()

        tools = await session.list_tools()
        print("Available tools:", [t.name for t in tools])

        result = await session.call_tool(
            "miko.factcheck",
            {"claim": "$BONK is partnering with Visa next week"}
        )
        print(result.content)

HTTP Transport

For deployments that cannot run a local MCP process (web-only environments, hosted agents), the MCP server is also available over Streamble HTTP:

https://mcp.mikoprotocol.com/mcp

Authentication via the same wallet-signed JWT as the stdio transport, passed in the Authorization: Bearer <jwt> header on each request to the endpoint.

Caching and Quota

The MCP server respects the same daily tier quota as the REST API. Identical tool calls within a 5-minute window return cached responses without counting against the daily quota.

Quota status is reported in tool call metadata:

{
  "_meta": {
    "tier": "Holder",
    "quota_remaining": 4,
    "cache_status": "HIT"
  }
}

What is MCP​

Server Package​

Authentication​

Tool Access by Tier​

Tools​

miko.factcheck​

miko.narrative​

miko.insights​

miko.narratives_trending​

miko.watchlist​

miko.persona​

Connecting from Claude Desktop​

Conversation example​

Connecting from Cursor​

Usage example​

Connecting from OpenAI Codex / Agents SDK​

Connecting from Gemini​

Connecting from Custom Agents​

HTTP Transport​

Caching and Quota​