agentcast-mcp
An MCP server that gives AI assistants theability to enforce structured output: extract JSON from messy LLM text, gateit against a shape spec, and produce the retry feedback message when the modelreturns the wrong shape.
Built on top of@mukundakatta/agentcast. Workswith Claude Desktop, Cursor, Cline, Windsurf, Zed, and any other MCP client.
Tools exposed
extract_json
Pull a JSON value out of messy LLM output. Tries the whole text, then afenced ```json ``` block, then the largest balanced {...} / [...]substring. Returns the parsed value plus which strategy succeeded.
{
"text": "Sure, here you go:\n```json\n{\"answer\": 42}\n```\nLet me know!"
}
→
{
"value": { "answer": 42 },
"found": true,
"source": "fenced_json"
}
source is one of whole, fenced_json, fenced_plain,balanced_substring, or none.
validate_response
Validate a parsed JSON value against an agentcast shape spec. Spec maps fieldname to type: string, number, boolean, array, object. Suffix with? for optional.
{
"value": { "name": "ada" },
"shape": { "name": "string", "age": "number" }
}
→
{
"valid": false,
"error": "missing required field 'age'"
}
build_retry_prompt
Given an attempt history, produce the validation-error feedback messageagentcast appends to the conversation when the model returned the wrongshape. Codifies the "validation error as feedback" pattern for non-Node MCPclients that want to drive the same retry loop manually.
{
"attempts": [
{ "text": "{\"name\":\"ada\"}", "error": "missing required field 'age'" }
],
"expected_shape": { "name": "string", "age": "number" }
}
→
{
"feedback": "Your previous response did not match the required shape. Error: missing required field 'age'\n\nTry again. Respond with ONLY valid JSON that fixes the error above.\n\nExpected shape: {\"name\":\"string\",\"age\":\"number\"}"
}
Install
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"agentcast": {
"command": "npx",
"args": ["-y", "@mukundakatta/agentcast-mcp"]
}
}
}
Cursor / Cline / Windsurf / Zed
Same shape, in the appropriate mcp.json for your client. Most clientsauto-discover via npx -y @mukundakatta/agentcast-mcp.
Local install
npm install -g @mukundakatta/agentcast-mcp
mcp-agentcast # listens on stdio
Why this matters
When an LLM is supposed to return structured data, it sometimes wraps theJSON in prose, fences, or hallucinated fields. Standard JSON.parse throws.Hand-rolled regex misses nested structure. This MCP server gives any modeldriving an agent a real handle on (1) pulling JSON out of the response,(2) checking it matches the expected shape, and (3) building the exact retryprompt that nudges the model to fix it on the next turn.
License
MIT.
