Connects Claude to a Postgres-backed behavioral oracle for Rust crate APIs with 66 million probe records across datetime, crypto, encoding, and JWT operations. Exposes five tools: signature search by type shape, behavior lookup for specific functions showing actual input/output tables, cross-implementation comparison tables, deprecation resolution, and routing for general API questions. No LLM generation in responses, just deterministic SQL and bge-m3 embedding retrieval. Useful when you need runtime behavior data instead of documentation claims, or want to see how chrono versus time actually handle the same duration string. Free tier at 10k requests per day, no signup.
Version-pinned behavioral oracle for Rust package APIs, built for AI coding
agents. Five deterministic tools backed by a Postgres substrate of
66,549,370 pattern_records and 107 successful probe runs across
7 task families (datetime, duration, base64, JSON strict-vs-permissive,
JWT verify, AEAD encrypt/decrypt, constant-time compare).
Hosted, free, no signup: https://api.codeitall.dev
This repository is codeitall's public surface — the API reference, OpenAPI spec, the "For AI Agents" page, and the per-family pages. The ingest/indexer and server source are not redistributed; you consume codeitall by querying the hosted API, not by running it yourself.
The differentiator from a docs RAG: the substrate captures runtime behavior (what a function actually does on a canonical input set, observed empirically), not just documentation (what it claims to do). It targets the documentation-vs-runtime gap — yank status, RustSec advisories, deprecation renames, and cross-implementation behavior tables.
| Tool | Use it for |
|---|---|
answer_api_question | Open-ended Rust API questions; routes deterministically (rule-based, no LLM) and falls back to bge-m3 cosine retrieval. |
signature_search | You know the signature shape (fn(&str) -> Result<Vec<u8>, _>); get ranked candidate functions + crate, downloads, yank status. |
behavior_lookup | A specific (crate, fn_name); get the probe-observation table verbatim — each row (input, outcome=ok|err|panic, value_or_error). |
compare_implementations | A task category (e.g. "base64 crates"); get the side-by-side cross-implementation behavior table. |
find_modern_equivalent | A suspected-deprecated call; get the current canonical alternative + yank/advisory rows. |
No generative LLM sits between a request and its response. Every response is SQL + deterministic templating; the only model in the hot path is bge-m3, a deterministic text→vector embedder used for the cosine fallback (same input → same vector → same hits). This is a regression-pinned invariant.
# Mint a free anonymous key (no email, no signup):
curl -sX POST https://api.codeitall.dev/v1/keys \
-H 'content-type: application/json' -d '{}'
# Ask a question (HTTP REST):
curl -sX POST https://api.codeitall.dev/v1/answer_api_question \
-H 'content-type: application/json' \
-d '{"question":"how do I parse a duration string in rust"}'
GET aliases exist for every tool (browse-only clients); signature_search
also accepts a shape_b64 (base64url) parameter so URL re-encoding can't
mangle type-signature characters.
https://api.codeitall.dev/mcp (Bearer key), or stdio
for self-hosters.POST/GET /v1/<tool>.openapi.json (live: https://api.codeitall.dev/openapi.json).for-ai-agents.html — positioning, tool list, key snippet (live: https://api.codeitall.dev/for-ai-agents.html).api-reference.html / api-reference.md — generated tool reference + scope/coverage.agents/ — one page per probe family.Free. Per-IP key issuance cap (10/day) and per-key quota (10,000/day). No paid tier, no email capture, no signup.
com.mcparmory/google-search
io.github.pipeworx-io/brave-search
marcopesani/mcp-server-serper
brave/brave-search-mcp-server
com.mcparmory/google-search-console
acamolese/google-search-console-mcp