Skip to content
GridRock
Get free key
Built for AI agents

The tools your LLM can call without you holding its hand.

OpenAPI 3.1 + MCP descriptor + RFC 7591 dynamic client registration + verbose canonical response shape. Bind once; works in Claude, OpenAI function-calling, LangGraph, AutoGen, MCP-compatible runtimes.

Tools published
14
Avg agent first-call
<5s
Token-cost discipline
Yes
Dynamic registration
RFC 7591

Endpoints

MethodPathCostTier
POST/oauth/register
Self-onboard, no human in the loop.
0free
POST/oauth/token
client_credentials → 1h JWT.
0free
GET/.well-known/mcp.json
Tool descriptor for MCP runtimes.
0free
GET/openapi.yaml
OpenAPI 3.1 with x-llm-* extensions.
0free

Try it from your terminal

No SDK install. Plain HTTP. Works the same from a Lambda, a Jupyter notebook, or an LLM tool call.

# 1. Register
curl -X POST https://api.gridrock.ai/oauth/register \
  -H 'Content-Type: application/json' \
  -d '{"client_name":"my-agent","scope":"read:geo read:hex"}'

# 2. Token
curl -X POST https://api.gridrock.ai/oauth/token \
  -d 'grant_type=client_credentials&client_id=...&client_secret=...&scope=read:geo'

# 3. Call
curl -H "Authorization: Bearer agtok_..." \
  "https://api.gridrock.ai/v1/intel/hex/8861aacd1bfffff"
For agents

Drop into Claude / OpenAI / LangGraph as gridrock

Bind once. Call on demand. Get back a verbose, deterministic shape that any LLM can reason over without prompt engineering.

// MCP server config
{
  "mcpServers": {
    "gridrock": {
      "url": "https://api.gridrock.ai/.well-known/mcp.json"
    }
  }
}
Why Bharat

Agents need predictable shapes. Western APIs assume a human in the loop with a debugger. Ours assume an LLM with a tool-call budget — every response carries an agent envelope so an LLM can act without re-prompting.

Related

Agents registry →MCP quickstart →Claude tool-use →

Start building with Agent Toolkit — free

No card. No sales call. 60-second self-serve, agents included.

Get a free key