self-inspect-mcp — Claude Code MCP server: install
This MCP server gives an agent a question about its own assumptions. Use it before a plan, decision, or final answer when the reasoning reaches a real block.
claude mcp add --transport http self-inspect-mcp https://api.ejentum.com/self-inspect-mcpWhat self-inspect-mcp does
self-inspect-mcp takes a description of the agent’s current thought or task and returns one metathought. The question redirects attention to assumptions, missing conditions, fixed decisions, or claims that do not follow. Its selector is deterministic and uses no LLM, embeddings, or API key.
The MCP interface exposes self_inspect with a thought argument. You can connect to the hosted HTTP endpoint or run the package locally with npx self-inspect-mcp; Python users also get dist/self_inspect.py, which uses the standard library. Call it at a genuine reasoning block. Running it after every step creates staged corrections and can make the result worse.
What it helps with
- check an assumption before taking action
- catch goal drift across a long chain
- challenge the first plausible answer
- surface missing conditions in a plan
Example prompts for the agent
- Before I continue, ask one question that tests the assumption behind this decision.
- I am about to lock in this architecture. Use
self_inspectto challenge my reasoning. - I am stuck at this step. Send the current thought to
self_inspectand use the returned question.
How to install self-inspect-mcp
1Platform CLI: remote server
claude mcp add --transport http self-inspect-mcp https://api.ejentum.com/self-inspect-mcp2Platform CLI
claude mcp add self-inspect-mcp -- npx -y self-inspect-mcp1Platform CLI: remote server
codex mcp add self-inspect-mcp --url https://api.ejentum.com/self-inspect-mcp2Platform CLI
codex mcp add self-inspect-mcp -- npx -y self-inspect-mcpConfig file
{
"mcpServers": {
"self-inspect-mcp": {
"url": "https://api.ejentum.com/self-inspect-mcp"
}
}
}{
"mcpServers": {
"self-inspect-mcp": {
"command": "npx",
"args": [
"-y",
"self-inspect-mcp"
]
}
}
}Config file
{
"servers": {
"self-inspect-mcp": {
"type": "http",
"url": "https://api.ejentum.com/self-inspect-mcp"
}
}
}{
"servers": {
"self-inspect-mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"self-inspect-mcp"
]
}
}
}1Platform CLI: remote server
gemini mcp add -t http self-inspect-mcp https://api.ejentum.com/self-inspect-mcp2Platform CLI
gemini mcp add self-inspect-mcp npx -y self-inspect-mcpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about self-inspect-mcp
How do I install self-inspect-mcp in Claude Code?
For the HTTP connection, run claude mcp add --transport http self-inspect-mcp https://api.ejentum.com/self-inspect-mcp.
What does self-inspect-mcp do?
It accepts an agent’s thought and returns one metathought that points at an assumption or gap. The result is a question for the agent to answer, not a solution.
Does self-inspect-mcp require an API key?
No. The hosted HTTP endpoint and the local npx self-inspect-mcp setup work without an API key.
