AILANG language reference, stdlib, examples and benchmarks for agents that write AILANG.
Scored across 23 tools
Most tools target a distinct resource+action pair, and the families (docs, examples, stdlib, benchmarks) are clearly separated by prefixes. The main soft spots are within-family: benchmarks_list vs benchmarks_for_model vs benchmark_run, and roadmap vs design_docs_list(state=planned), which could occasionally be confused. Descriptions generally resolve the residual ambiguity.
The dominant convention is noun_verb (example_get, stdlib_search, limitations_list, benchmark_run), but it is broken by bare-noun tools (roadmap, docs_nav, install_guide, onboarding_guide, design_doc) and a verb_noun outlier (submit_feedback). Singular/plural is semi-predictable (singular = get one, plural = list) yet still inconsistent in places like design_doc vs design_docs_list and example_get vs examples_list.
23 tools is on the heavy edge of the recommended band. The domain is genuinely broad (versions, changelog, design docs, docs nav/search, examples, effects, stdlib, benchmarks, install, onboarding, limitations, prompts, roadmap, feedback), so most tools earn their place, but the surface feels like it accumulated rather than being scoped.
Coverage across docs, examples, stdlib, releases and benchmarks is broad, but there is a notable dead end: docs_search returns only excerpts and headings, and the description claims the full page is available via docs_nav or example_get, yet no doc_get/fetch-page tool exists. Benchmark surface also lacks a way to enumerate categories or suites independently of runs, leaving some workflows partially closed.
List all AILANG versions the server can answer for.
Inputs: no named parameters.
One run's full result (code, compile_ok, runtime_ok, stdout_ok, cost, tokens).
Inputs: runId (required).
Pairwise model comparison. Output: {a_only_passes, b_only_passes, both, neither}
Inputs: modelA (required), modelB (required), version (required).
Where does a given model pass/fail? Optional version filter. Output: {passes: [...], fails_by_category: {...}, total_cost}
Inputs: model (required), version (required).
List benchmark suites and runs, optionally filtered to a since_version cutoff.
Inputs: sinceVersion (required).
What changed in a given release (added/changed/fixed/breaking).
Inputs: version (required).
Full design doc by slug (e.g. "m-agent-mcp-website").
Inputs: slug (required).
List planned/implemented/rejected design docs, optionally filtered by version. state = "planned" | "implemented" | "rejected" | ""; empty = all.
Inputs: state (required), version (required).
Full-text search across doc Markdown for the given version. Filters server-side on lowercased body + title; returns the top 10 matches, docs pages before archived teaching prompts (prompts/v*.md match almost any query), each as path, title, its first 10 headings and an excerpt around the match: the full page is a docs_nav / example_get away. Whole bodies made one search ~30 KB, too much for a tool result in a listed connector.
Inputs: query (required), forVersion (required).
All declared effects with capability mapping and the version they appeared in.
Inputs: forVersion (required).
Best-fit examples for a free-form concept query ("how do I use effects?"). Filters server-side on why + title; returns up to 5 matches.
Inputs: concept (required), forVersion (required).
Full source + metadata for one example by path.
Inputs: path (required), forVersion (required).
Filter examples that pass under for_version, optionally by features/status. features = comma-separated tags ("effects,records"); empty = no filter. status = "passing" | "failing" | ""; empty = passing only.
Inputs: status (required), features (required), forVersion (required).
Harness-aware install instructions for AILANG. harness: "claude-code" | "gemini-cli" | "codex-cli" | "manual" | "build-from-source" | "" platform: only meaningful for "manual" (e.g. "darwin-arm64", "linux-amd64"); ignored otherwise Empty harness returns ALL harnesses + the remote_mcp config block.
Inputs: harness (required), platform (required).
Known limitations (Y-combinator, etc.) optionally filtered by category.
Inputs: category (required), forVersion (required).
Stepwise onboarding flow for using AILANG. role: "agent" (AI coding agent flow, default) | "developer" (human dev flow) | "" (returns both) For v1 returns the full guide; client picks the relevant role. v2 may filter on AILANG side once we have proper record-projection helpers.
Inputs: role (required).
Fetch the canonical teaching prompt for a version. kind = "agent" | "devtools" | "compact" (full prompt = empty).
Inputs: kind (required), forVersion (required).
What's shipping next, what's been rejected and why.
Inputs: no named parameters.
Full module: exports + signatures + docstrings.
Inputs: name (required), forVersion (required).
List stdlib modules available in this AILANG version with one-line summaries.
Inputs: forVersion (required).
Find stdlib functions by name, signature, or keyword. Filters the pre-built search index server-side; returns top 20 matches.
Inputs: query (required), forVersion (required).
Anonymous bug report / feature request / docs gap, queued for human review. Default routing: public-feedback inbox (general AILANG). Pass package="vendor/name" (e.g. "sunholo/auth") to route to that package's pkg:vendor/name inbox where its autonomous agent watches. Categories: bug, feature, docs, limitation. Body limit 10KB, snippet limit 4KB. Optional contact field for follow-up; opaque to the server. Set auto_dispatch=true to authorize the package agent to act on your submission immediately (default false — files for human triage; pkg-feedback agent template lands in a separate sprint).
Inputs: body (required), title (required), category (required), ailang_version (required), +4 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
AI-native programming language and toolchain for accountable, verifiable agent-driven development.
Write, type-check and run AILANG in a sandbox, with Z3-proved contracts and docs search.
Machine-readable utilities and datasets for AI agents.
Persistent memory, hybrid search and a goal graph for AI agents, over stdio or remote HTTP.
Add one secure layer between your agents and this server.