falksterheidi
Join waitlist →
Developers · MCP

Your company’s brain is an MCP server

MCP is how AI tools talk to the world. Heidi speaks it natively: one hosted endpoint, streamable HTTP, protocol 2025-06-18. Point any MCP client at it and that client can search your memory, look up entities, and ask Heidi anything.

Why this matters

Every AI tool you use starts from zero. Your coding agent does not know why the payment service is split in two. Your research agent does not know which customers matter. Connect them to Heidi and they all read from the same brain: the decisions, the incidents, the people, the promises. One memory, every tool.

Connect your client

Claude Code

terminal
claude mcp add --transport http heidi https://mcp.falkster.ai/mcp \
  --header "Authorization: Bearer <your Heidi API token>"

Claude Desktop

claude_desktop_config.json
// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "heidi": {
      "type": "http",
      "url": "https://mcp.falkster.ai/mcp",
      "headers": {
        "Authorization": "Bearer <your Heidi API token>"
      }
    }
  }
}

Cursor

mcp.json
// ~/.cursor/mcp.json
{
  "mcpServers": {
    "heidi": {
      "url": "https://mcp.falkster.ai/mcp",
      "headers": {
        "Authorization": "Bearer <your Heidi API token>"
      }
    }
  }
}

Windsurf

mcp_config.json
// ~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "heidi": {
      "serverUrl": "https://mcp.falkster.ai/mcp",
      "headers": {
        "Authorization": "Bearer <your Heidi API token>"
      }
    }
  }
}

Codex CLI

config.toml
# ~/.codex/config.toml
[mcp_servers.heidi]
url = "https://mcp.falkster.ai/mcp"
authorization = "Bearer <your Heidi API token>"

Any MCP client

raw JSON-RPC
curl https://mcp.falkster.ai/mcp \
  -H "Authorization: Bearer $HEIDI_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Perplexity Comet and ChatGPT work too: Comet adds the server under Sources & Tools, ChatGPT wraps the REST surface as a custom GPT Action. Your instance serves paste-ready blocks for every client at /api/v1/mcp/client-config/{surface}.

What your client gets

tools/listreturns Heidi’s registry: the ask_heidi tool that runs a full agent turn, plus the memory tools your agent will actually live on: memory_search, memory_recall, memory_get_entity, memory_list_facts, and friends. Every answer carries its source.

And what it cannot do

Scopes apply to MCP exactly like REST. A default token can ask and read; it cannot send email, post to Slack, or touch anything side-effecting. Those run through the agent path with confirmation gating. Your intern’s experiment cannot page your customers.

Two minutes from token to connected

Join the waitlist for your instance, mint a token in the Developer Portal, paste one block. Done.

See the Claude Code guide
Answers →
Ask AI about Heidi:
Share this: