M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
A

AI Native Lang (AINL)

Community
Turn AI conversations into verifiable, repeatable structured workflows.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #60 of 438
Stars
★ 650 Very popular
Transport
stdio (local process)
Runtime
Python 3.10+
Credentials
No credential needed
License
Apache-2.0
Last commit
Tools
13
64FMRS · C

This is an Apache-2.0 open-source AINL repository maintained by sbhooley. It provides a compiler, deterministic runtime, graph IR, CLI, HTTP runner, and MCP server. The README documents MCP tools, stdio usage, and optional adapters; exact permissions depend on the workflow and runtime configuration.

Strongest · Documentation 15/20 Weakest · Reliability 10/20

Reliability
10/20
Security and permissions
11/20
Maintenance
15/20
Documentation
15/20
Setup experience
13/20
Why each score
Reliability 10/20
The README describes ainl-mcp, multiple tools, exposure controls, limits, diagnostics, and numerous test and CI entry points, providing evidence of a reviewable intended path. However, the server manifest is not cached, so declared tools cannot be checked against implementation; no MCP initialization, tool-list handshake, or tests were executed, and the supplied README is truncated at the end. This supports a plausible happy path but warrants deductions for missing verifiable execution evidence.
Security and permissions 11/20
The README identifies core-only default adapters, exposure profiles, capability grants, run limits, HTTP/A2A host controls, redirects off by default, SSRF options, and a security report, showing some least-privilege structure. It also exposes potentially high-impact HTTP, filesystem, SQLite, queue, external-LLM, A2A, Solana, payment, and installation capabilities. The source provided does not establish the actual defaults, complete parameters, confirmation behavior, credential flows, or manifest-level boundaries, so substantial points are deducted. No explicit malicious behavior or real secret was found that would trigger a security red line.
Maintenance 15/20
The repository is stated to be non-archived, uses Apache-2.0, and documents release surfaces through versions 1.3.4–1.8.0, release hygiene, contribution paths, SECURITY.md, and vulnerability-reporting areas, with identifiable ownership. There is no supplied evidence of commit cadence, dependency-update practice, issue-response performance, or a concrete security-response process, so the score is below full marks.
Documentation 15/20
The README supplies layered installation paths, CLI/HTTP/MCP usage, host integrations, limits, exposure profiles, security concepts, adapters, examples, troubleshooting entry points, benchmark methodology, and disclosed limitations. Points are deducted because the manifest is missing, complete MCP tool parameters and actual behavior cannot be independently reconciled, several performance and product claims depend on files not supplied here, and the provided README excerpt is incomplete.
Setup experience 13/20
The source provides pip/pipx installation, ainl setup --auto, stdio startup instructions, a generic MCP configuration fallback, and host-specific commands, making the intended setup fairly clear. Points are deducted because Python 3.10+ is required, host integrations still vary in configuration and additional steps, the manifest is unavailable, and installation or connection was not verified in this static review.

Static review · not runListed 2026-07-29

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the networkConnects to a database

Best for

  • Teams building multi-step AI workflows with state, memory, and tool use
  • Teams seeking to reduce repeated prompt-loop orchestration
  • Projects requiring strict compile-time validation or execution auditing
  • Users of Claude Code, Codex, or other MCP-compatible agents

Not for

  • Users who only need simple hand-written scripts
  • Teams that already have deterministic runners and use LLMs only at judgment gates
  • Projects with no need for workflow validation, auditing, or MCP integration

Required permissions

  • Install and run a local Python program
  • Access workspace files, cache, or SQLite when corresponding adapters are enabled
  • Access networks or external services when HTTP, A2A, or other adapters are enabled
  • Use relevant environment configuration when workflows require secrets or external services

Risks and side effects

  • Enabling HTTP or A2A adapters introduces external network access risk
  • Workflows may access files, caches, databases, or secrets depending on adapters and policy configuration
  • Misconfigured adapter, step, runtime, or workspace limits may broaden execution scope
  • Machine-payment profiles are opt-in but may process HTTP 402 payment flows when enabled

Setup

Before you start

Runtime：Python 3.10+

Other optional settings (5)
AINL_MCP_EXPOSURE_PROFILE optional Optional; scopes which MCP tools/resources are exposed (e.g. validate_only, inspect_only, safe_workflow, full)
AINL_DRY_RUN optional Optional; set to 1 to run in dry-run mode without real side effects such as transactions
AINL_ALLOW_IR_DECLARED_ADAPTERS optional Optional; relaxes the AINL_HOST_ADAPTER_ALLOWLIST host adapter policy
AINL_HOST_ADAPTER_ALLOWLIST optional Optional; controls the host adapter allowlist
AINL_EFFICIENT_MODE optional Optional; env signal to the ArmaraOS kernel for efficient mode

Requires Python 3.10 or newer. Install the MCP extra with pipx install 'ainativelang[mcp]' && ainl setup --auto, or use python3 -m pip install --user 'ainativelang[mcp]' && ainl setup --auto. Start ainl-mcp and configure it as a stdio server in the MCP client. Run ainl doctor to check integration.

Check that it works

After installing 'ainativelang[mcp]' and running 'ainl setup --auto', the tools ainl_validate, ainl_compile, and ainl_run should appear in your MCP client's tool list; calling ainl_capabilities and getting a capabilities response confirms the connection works.

Troubleshooting

  1. Confirm Python 3.10 or newer
  2. Confirm that ainativelang[mcp] is installed rather than only the base package
  3. Run ainl doctor to check host integration
  4. Check the MCP exposure profile, adapter allowlist, and execution limits
  5. If auto-cache is enabled, confirm that cache.json contains valid JSON
  6. For the dashboard, confirm that tests/emits/server/server.py has been generated

Things to try

Once connected, you can ask your AI assistant things like:

  • Validate my .ainl workflow file strictly with ainl_validate
  • Compile main.ainl with ainl_compile and show me the IR
  • Run this workflow with ainl_run and return the execution result
  • Use ainl_get_started to walk me through writing my first AINL workflow

Tools 13

ainl_validate read-only
Validates an AINL workflow and returns structured diagnostics.
ainl_compile read-only
Compiles an AINL workflow into a deterministic graph intermediate representation.
ainl_capabilities read-only
Reports available capabilities.
ainl_security_report read-only
Reports security configuration and capability information.
ainl_run writes
Executes an AINL workflow.
ainl_step_examples read-only
Provides AINL step examples.
ainl_get_started read-only
Provides onboarding guidance and wizard state.
ainl_ir_diff read-only
Compares two AINL intermediate representations.
Show 5 more tools
ainl_fitness_report read-only
Generates a workflow fitness report from multiple runs.
ainl_list_ecosystem read-only
Lists ecosystem workflow and agent resources.
ainl_import_clawflow writes
Imports Clawflows-style workflows.
ainl_import_agency_agent writes
Imports Agency-Agents-style agents.
ainl_import_markdown writes
Imports Markdown as an AINL workflow.

Use cases

Authoring and validating deterministic AI workflows
Compiling workflows into graph IR for repeated execution
Providing workflow validation, compilation, and execution to coding agents through MCP
Importing Clawflows- or Agency-Agents-style content
Building workflows with HTTP, filesystem, cache, SQLite, LLM, and memory adapters

Supported clients

Claude Code
Codex
Gemini CLI
Claude DesktopPartial support

Listed from the project's documentation, not tested by this site.

Overview

AINL is a compact, graph-canonical AI-native workflow language and runtime. It compiles workflows into a deterministic intermediate-representation graph and supports multi-step execution, state and memory, tool use, validation, control, visualization, an HTTP runner, and MCP integration.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision bbdc784aa432 Data synced 2026-10-19 Read the FMRS scoring method