Build with Engrammic

Endpoint

Beta MCP endpoint: https://beta.engrammic.ai/mcp/

Authentication

Engrammic uses OAuth. Your agent triggers a browser-based flow:

Agent calls the MCP endpoint without a token
Server returns an OAuth URL
User opens URL in browser, authenticates
Agent receives token via callback
Subsequent calls include the token

Token refresh is automatic. If a token expires mid-session, the server returns AUTH_REQUIRED and you re-trigger the flow.

MCP Client Setup

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";

const transport = new StreamableHTTPClientTransport(
  new URL("https://beta.engrammic.ai/mcp/")
);

const client = new Client({
  name: "my-agent",
  version: "1.0.0"
});

await client.connect(transport);

// Store an observation
await client.callTool("remember", {
  content: "User prefers dark mode",
  context: "stated in settings discussion"
});

// Retrieve knowledge
const result = await client.callTool("recall", {
  query: "user preferences"
});

from mcp import Client
from mcp.client.streamable_http import StreamableHTTPClientTransport
import asyncio

async def main():
    transport = StreamableHTTPClientTransport("https://beta.engrammic.ai/mcp/")
    
    client = Client("my-agent", "1.0.0")
    await client.connect(transport)
    
    # Store an observation
    await client.call_tool("remember", {
        "content": "User prefers dark mode",
        "context": "stated in settings discussion"
    })
    
    # Retrieve knowledge
    result = await client.call_tool("recall", {
        "query": "user preferences"
    })

asyncio.run(main())

Core Operations

Store an observation:

{"tool": "remember", "args": {"content": "...", "context": "..."}}

Store a claim with evidence:

{"tool": "learn", "args": {"claim": "...", "evidence": [{"type": "direct_statement", "content": "..."}]}}

Retrieve knowledge:

{"tool": "recall", "args": {"query": "..."}}

Trace provenance:

{"tool": "trace", "args": {"query": "why do you believe X?"}}

Error Handling

Errors return:

{"error": {"code": "ERROR_CODE", "message": "Human-readable message"}}

Code	Meaning	Action
`AUTH_REQUIRED`	Token missing or expired	Re-trigger OAuth flow
`INVALID_TOKEN`	Token rejected	Re-authenticate
`SILO_NOT_FOUND`	Silo doesn't exist	Check silo_id config
`NODE_NOT_FOUND`	Referenced node missing	Node was deleted or ID is wrong
`RATE_LIMITED`	Too many requests	Back off, retry with exponential backoff
`VALIDATION_ERROR`	Bad parameters	Check required fields

Rate Limits

Beta limits (subject to change):

100 requests per minute per silo
10,000 nodes per silo

Coming Soon

Native TypeScript and Python SDKs are in development. For now, use the MCP SDK directly as shown above.