Huntadocs

Connect over MCP

Point any MCP client at your memory. One URL, one bearer token.

The Hunta memory engine is an MCP server (Streamable HTTP) at https://mcp.hunta.ai/mcp. Any MCP client connects with a bearer token; the tenant is bound to the token, never passed as a parameter.

Generic MCP client

Most clients accept an mcpServers block:

{
  "mcpServers": {
    "gather": {
      "url": "https://mcp.hunta.ai/mcp",
      "headers": { "Authorization": "Bearer <your-agent-key>" }
    }
  }
}

Codex (TOML)

[mcp_servers.gather]
url = "https://mcp.hunta.ai/mcp"

[mcp_servers.gather.http_headers]
Authorization = "Bearer <your-agent-key>"

What your agent gets

The server exposes memory as branded tools your agent calls by name:

  • gather writes a memory (and capture proposes one into the review queue)
  • recall and search read sealed facts back, with provenance and point-in-time (as_of)
  • facts lists the current canon
  • the Instinct reflex set, plus curation actions (promote, reject) with the right scopes

These are the same three verbs as the REST API (Gather, Recall, Instinct); see the API reference. Scopes on the token decide what a given agent can do:

ScopeAllows
memory:readrecall and search, read the attestation report
memory:capturepropose candidates into the staging queue
memory:writedirect writes to canon (owner tooling)
memory:curatepromote or reject candidates

Give harness agents capture plus read. Keep write and curate for owners and trusted tooling. The backend enforces scopes server-side, so a mis-scoped client cannot exceed them.

REST instead of MCP

Everything is also plain REST on the same host. See the API reference.

On this page