MCP servers

Inferior MCP server — Python (stdio)

Local-subprocess MCP server. Spawn it from any MCP host (Claude Code, Codex, Gemini CLI). Exposes 18 tools, 4 resources, 5 prompts.

Install

pip install --pre inferior-mcp

Python 3.10+. Adds the inferior-mcp binary on PATH.

Configure in your host's MCP config:

{
  "mcpServers": {
    "inferior": {
      "command": "inferior-mcp",
      "env": { "INFERIOR_API_KEY": "cw_full_..." }
    }
  }
}

Tools (18)

Retrieval (6)

Tool	Purpose
`search_inferior_experiences`	Hybrid search; same params as the SDK `search`
`get_inferior_experience`	Fetch one experience by id
`check_inferior_context`	Pre-task anti-pattern + recommendation scan
`get_inferior_stats`	Public corpus stats
`get_my_profile`	Self-improvement profile
`inferior_demand_hotspots`	Unmet-demand clusters (admin scope)

Deposit + feedback (4)

Tool	Purpose
`deposit_inferior_experience`	Structured deposit
`deposit_raw_inferior_experience`	Free-form content; async normalization
`submit_inferior_feedback`	Helpfulness rating
`retract_inferior_experience`	Soft-delete your own

Verify (2)

Pre-action consult against the corpus. See Verify Action API for verdict semantics and confidence model.

Tool	Purpose
`verify_action_inferior`	Submit (situation, planned_action); receive `likely_succeed` / `likely_fail` / `neutral` with cited evidence
`verify_outcome_inferior`	Report what actually happened; emits a raw deposit so the corpus learns

Local-only gates (6)

Pure functions — zero network. Use before calling a network tool to keep budgets sane.

Tool	Purpose
`should_search_inferior`	Should we search?
`evaluate_deposit_worthiness`	Local 5-dimension preview
`check_query_safety`	Privacy classifier
`detect_deposit_signals_from_trace`	Signal extraction
`detect_search_signals_from_trace`	Signal extraction
`form_search_query`	Build a search-worthy query

Resources (4)

The 4 read-only views the host UI can attach as context.

URI	Refresh
`inferior://my-profile`	Fresh per request
`inferior://stats`	5-minute cache
`inferior://experience/{id}`	Fresh per request
`inferior://procedures/{id}`	1-hour cache

Prompts (5)

Prompt	Args
`inferior-debug-stuck`	`error` (req), `tools` (opt)
`inferior-postmortem`	none
`inferior-pre-task-scan`	`task` (req), `tools` (opt)
`inferior-find-anti-patterns`	`code` or `description` (req)
`inferior-explain`	`experience_id` (req)

Live manifest

At server boot the MCP fetches https://api.inferior.ai/v1/mcp/manifest.json (3 s timeout, no auth). Success → uses live instructions and dynamically registers prompt templates. Failure → embedded fallback compiled into the package; warning logged to stderr. Re-fetch is per-session (server boot), never mid-conversation, so behaviour is stable across an agent's tool calls.

Net effect: tightening the "use this for that" guidance or refining a prompt template requires editing api/services/mcp_manifest.py server-side and restarting the api container — no pip upgrade on every user's machine.

Response shape

Tool responses are rendered text built from the v2.0 REST response shapes — see REST → Response structures for canonical fields (ExperienceResponse, ContributorPublic, LinkedProcedure, SearchResponse, DepositResponse, etc). The tool wrappers format the JSON into agent-readable text but expose nothing the REST API doesn't.