Connect Haycion via MCP

Last updated: 2026-09-16

Haycion MCP exposes an MCP server so AI assistants (Claude Desktop, Claude Code, any MCP-compatible client) can use it directly from a chat. Research companies and competitors with an AI advisor, and run vendor comparisons.

• Server URL: https://mcp.haycion.com/api/mcp
• Transport: Streamable HTTP
• Auth: OAuth 2.1 with PKCE — most MCP clients handle the whole flow automatically the first time you add the server.

Connect

1. In your MCP client, add a new remote server pointing at https://mcp.haycion.com/api/mcp.
2. Your client will discover the OAuth endpoints from https://mcp.haycion.com/.well-known/oauth-authorization-server and https://mcp.haycion.com/.well-known/oauth-protected-resource/api/mcp, then open your browser to sign in and approve access.
3. Approve the requested scopes on the consent screen.
4. Your client exchanges the authorization code for an access token and is ready to use the tools below.

Claude Desktop / Claude Code example

{
  "mcpServers": {
    "haycion-mcp": {
      "url": "https://mcp.haycion.com/api/mcp"
    }
  }
}

Scopes

Scope	Grants
kg:agents:read	Agents (Read)
kg:agents:manage	Agents (Manage)
kg:agents:advise	Agents (Advise)
kg:vendor-research:run	Vendor Research (Run)

Agents (Read)

list_agents

List agents — scope: kg:agents:read (Read)

Lists every agent (KnowledgeGraph workspace) the authenticated user is a member of.

Input schema:

{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}

get_agent

Get agent — scope: kg:agents:read (Read)

Gets one agent's full profile (organization, products, ICPs, market segments, value proposition) by its workspace id.

Input schema:

{
  "type": "object",
  "properties": {
    "workspaceId": {
      "type": "string",
      "description": "The agent's workspace id, from list_agents."
    }
  },
  "required": [
    "workspaceId"
  ],
  "additionalProperties": false
}

Agents (Manage)

create_agent

Create agent — scope: kg:agents:manage (Manage)

Creates a new agent by crawling its company website and/or a short brief, kicking off the async research build. Provide at least one of websiteURL or orgBrief.

Input schema:

{
  "type": "object",
  "properties": {
    "websiteURL": {
      "description": "The company's website URL, if known.",
      "type": "string"
    },
    "orgBrief": {
      "description": "A short brief describing the company, used when there's no website to crawl.",
      "type": "string"
    }
  },
  "additionalProperties": false
}

Agents (Advise)

query_agent

Query agent — scope: kg:agents:advise (Advise)

Asks a free-text question about one of your agents and gets an LLM-synthesized answer grounded strictly in that agent's already-researched profile (organization, products, ICPs, market segments, value proposition). Does not run new research or answer questions outside what's already been built for this agent — use create_agent/wait for research to finish first if the profile is thin. Stateful: every call continues the same ongoing conversation for that agent, so you can ask follow-up questions naturally. Use this instead of get_agent when you want an answer rather than raw structured data.

Input schema:

{
  "type": "object",
  "properties": {
    "workspaceId": {
      "type": "string",
      "description": "The agent's workspace id, from list_agents."
    },
    "message": {
      "type": "string",
      "description": "A free-text question or follow-up message about this agent's business."
    }
  },
  "required": [
    "workspaceId",
    "message"
  ],
  "additionalProperties": false
}

get_agent_advisor

Get agent Advisor history — scope: kg:agents:advise (Advise)

Gets the full message history of one agent's Advisor conversation (see query_agent), oldest first.

Input schema:

{
  "type": "object",
  "properties": {
    "workspaceId": {
      "type": "string",
      "description": "The agent's workspace id, from list_agents."
    }
  },
  "required": [
    "workspaceId"
  ],
  "additionalProperties": false
}

clear_agent_advisor

Clear agent Advisor history — scope: kg:agents:advise (Advise)

Permanently deletes one agent's whole Advisor conversation history (does not touch the agent's own research data). Irreversible.

Input schema:

{
  "type": "object",
  "properties": {
    "workspaceId": {
      "type": "string",
      "description": "The agent's workspace id, from list_agents."
    }
  },
  "required": [
    "workspaceId"
  ],
  "additionalProperties": false
}

Vendor Research (Run)

start_vendor_research

Start vendor research — scope: kg:vendor-research:run (Run)

Kicks off a vendor comparison research job for one of the user's companies, scoring candidate vendors against a requirements brief. Always runs as a buyer-side evaluation.

Input schema:

{
  "type": "object",
  "properties": {
    "workspaceId": {
      "type": "string"
    },
    "title": {
      "type": "string"
    },
    "requirementsBrief": {
      "type": "string"
    },
    "vendorURLs": {
      "description": "Known candidate vendor URLs to include, if any.",
      "type": "array",
      "items": {
        "type": "string"
      }
    },
    "useCache": {
      "type": "boolean"
    },
    "includeSelf": {
      "description": "Whether to include the user's own company as a comparison candidate.",
      "type": "boolean"
    }
  },
  "required": [
    "workspaceId",
    "title",
    "requirementsBrief"
  ],
  "additionalProperties": false
}

Utility

whoami

Returns the identity (user, OAuth client, and granted scopes) the calling MCP client is currently authenticated as. Always available regardless of granted scopes.