Aviary MCP Server

MCP server that exposes NASA's Aviary aircraft design optimizer as 9 tools over streamable-HTTP. An AI agent (or any MCP client) can explore the design space, modify aircraft parameters, run gradient-based trajectory optimization, and retrieve results — all through standard MCP calls.

Related Repositories

• MAS-Aviary — Multi-agent LLM framework that uses this MCP server as its backend for aircraft design optimization. MAS-Aviary orchestrates specialized agents (exploration, modification, simulation, analysis) that collaborate to find fuel-optimal aircraft configurations by calling the tools exposed here.

Quick start

1. Create the conda environment

Aviary 0.9.10 requires specific versions of OpenMDAO and Dymos. Newer versions introduce a unit-compatibility bug that breaks trajectory optimization. The pinned versions below are the only combination known to work.

conda create -n aviary python=3.11 -y
conda activate aviary
pip install -r requirements.txt

requirements.txt pins:

Package	Version	Why
aviary	0.9.10	NASA aircraft design/optimization
openmdao	3.36.0	MDO framework — 3.41+ breaks aviary unit handling
dymos	1.13.1	Trajectory optimization — 1.15+ breaks aviary unit handling
fastmcp	>=3.0.0	MCP server framework
pytest	latest	Test runner

2. Start the server

conda activate aviary
python aviary_mcp_server.py

The server listens on http://localhost:8600/mcp (streamable-HTTP transport).

3. Connect a client

Point any MCP-compatible client at http://localhost:8600/mcp. The server exposes 9 tools that follow a session-based lifecycle.

MAS-Aviary integration: The MAS-Aviary multi-agent framework connects to this endpoint automatically. See the MAS-Aviary quickstart guide for instructions on running the full multi-agent optimization pipeline.

get_design_space  →  create_session  →  set_aircraft_parameters  →  configure_mission
                                              ↓
                     check_constraints  ←  get_results / get_trajectory  ←  run_simulation
                                              ↑
                                     validate_parameters (optional pre-check)

4. Verify the installation

# Unit tests (no server required)
python -m pytest tests/test_aviary_mcp_tools.py -v -m "not slow"

# End-to-end connection test (server must be running)
python test_aviary_mcp_connection.py

Tools

#	Tool	Description
1	get_design_space	List 10 modifiable aircraft parameters with defaults, units, and bounds
2	create_session	Start a new session (loads 737/A320-class baseline aircraft)
3	set_aircraft_parameters	Modify wing, fuselage, or engine parameters with validation
4	configure_mission	Set range, passengers, cruise Mach, altitude, optimizer iterations (defaults: 1500 nmi, 162 pax, Mach 0.785, FL350)
5	validate_parameters	Static checks + quick model evaluation (~6-9s) without running the optimizer
6	run_simulation	Run SLSQP trajectory optimization (default 200 iterations, ~40-60s)
7	get_results	Read fuel burn, GTOW, wing mass, reserve fuel, zero-fuel weight
8	get_trajectory	Extract timeseries arrays: altitude, Mach, mass, throttle, drag, distance vs time
9	check_constraints	Evaluate pass/fail against user-defined limits (<=, >=, ==)

Design parameters

Parameter	Default	Units	Range
Aircraft.Wing.ASPECT_RATIO	11.22	—	7–14
Aircraft.Wing.AREA	124.6	m²	100–160
Aircraft.Wing.SPAN	37.35	m	28–48 (derived: span = √(AR × area))
Aircraft.Wing.SWEEP	25.0	deg	15–40
Aircraft.Wing.TAPER_RATIO	0.278	—	0.15–0.45
Aircraft.Fuselage.LENGTH	37.79	m	28–50
Aircraft.Fuselage.MAX_HEIGHT	4.06	m	3–5.5
Aircraft.Fuselage.MAX_WIDTH	3.76	m	3–5.5
Aircraft.Engine.SCALED_SLS_THRUST	28928	lbf	20k–45k (read-only, derived from SCALE_FACTOR)
Aircraft.Engine.SCALE_FACTOR	1.0	—	0.8–1.5

Wing area, span, and aspect ratio are coupled: AR = span² / area. Span is a derived parameter — it is computed from aspect ratio and area and should not be set directly. If you set span explicitly, the server will warn if the coupling constraint is violated by more than 5%. Set ASPECT_RATIO and AREA instead, and span will be derived automatically. SCALED_SLS_THRUST is similarly derived from SCALE_FACTOR.

Mission defaults

Parameter	Default
Range	1,500 nmi
Passengers	162
Cruise Mach	0.785
Cruise altitude	FL350 (35,000 ft)
Optimizer iterations	200 (SLSQP)

Project structure

aviary_mcp_server.py        MCP server — 9 tool definitions
aviary_runner.py             Aviary Level 2 API wrapper
session_manager.py           Session lifecycle, 30-min idle timeout
design_space.py              Parameter metadata and validation
requirements.txt             Pinned dependencies
start_aviary_server.py       Convenience start script
tests/                       pytest suite (25 unit + 31 integration)
test_aviary_mcp_connection.py  End-to-end smoke test
run_reference_benchmark.py   Standalone benchmark (~7,001 kg fuel baseline)
CHANGELOG.md                 Version history
RESULTS.md                   Batch evaluation of 8 agent configurations
docs/index.html              Project landing page
figures/                     Generated plots

Troubleshooting

unitless vs 1 unit error during optimization: You have a newer version of openmdao or dymos installed. Downgrade to the pinned versions:

pip install openmdao==3.36.0 dymos==1.13.1

build_model() not found: This method does not exist in Aviary 0.9.10. The server uses the correct call sequence internally — if you see this error, your aviary version is wrong.

Simulation hangs or times out: run_simulation accepts a timeout_seconds parameter (default 300). Complex configurations may need more time. Check that the optimizer iteration count (configure_mission → optimizer_max_iter) is reasonable.

Citation

If you use this MCP server in your research, please cite:

@inproceedings{ezemba2026structure,
  title={Structure Shapes Coordination: An Empirical Evaluation of Organizational
         Design Principles in Multi-Agent Engineering Design},
  author={Ezemba, Jessica and McComb, Christopher and Tucker, Conrad},
  booktitle={ASME 2026 International Design Engineering Technical Conferences
             and Computers and Information in Engineering Conference (IDETC/CIE2026)},
  address={Houston, TX},
  month={August 23--26},
  year={2026}
}

This server wraps NASA's OpenMDAO/Aviary:

@software{aviary,
  title={Aviary: A Multidisciplinary Design, Analysis, and Optimization Tool
         for Aircraft},
  author={{NASA Glenn Research Center}},
  url={https://github.com/OpenMDAO/Aviary},
  year={2024}
}