Context : Clouds are ubiquitous in exoplanet atmospheres and represent a challenge for the model interpretation of their spectra .
Complex cloud models are too numerically costly for generating a large number of spectra , while more efficient models may be too strongly simplified .

Aims : We aim to constrain the atmospheric properties of the directly imaged planet HR 8799e with a free retrieval approach .

Methods : We use our radiative transfer code petitRADTRANS for generating spectra , which we couple to the PyMultiNest tool .
We added the effect of multiple scattering which is important for treating clouds .
Two cloud model parameterizations are tested : the first incorporates the mixing and settling of condensates , the second simply parameterizes the functional form of the opacity .

Results : In mock retrievals , using an inadequate cloud model may result in atmospheres that are more isothermal and less cloudy than the input .
Applying our framework on observations of HR 8799e made with the GPI , SPHERE and GRAVITY , we find a cloudy atmosphere governed by disequilibrium chemistry , confirming previous analyses .
We retrieve that { C / O } = 0.60 _ { -0.08 } ^ { +0.07 } .
Other models have not yet produced a well constrained C/O value for this planet .
The retrieved C/O values of both cloud models are consistent , while leading to different atmospheric structures : cloudy , or more isothermal and less cloudy .
Fitting the observations with the self-consistent Exo-REM model leads to comparable results , while not constraining C/O .

Conclusions : With data from the most sensitive instruments , retrieval analyses of directly imaged planets are possible .
The inferred C/O ratio of HR 8799e is independent of the cloud model and thus appears to be a robust .
This C/O is consistent with stellar , which could indicate that the HR 8799e formed outside the CO _ { 2 } or CO iceline .
As it is the innermost planet of the system , this constraint could apply to all HR 8799 planets .